feat: add sidebar updater for comfyui_o1key

Add Grok and VEO video workflow support
Update image and video workflow nodes
2026-09-24 17:51:05 +08:00 · 2026-06-04 16:06:23 +08:00 · 2026-05-28 16:44:30 +08:00 · 2026-05-26 18:05:07 +08:00 · 2026-05-26 00:57:46 +08:00 · 2026-05-25 18:33:15 +08:00
94 changed files with 27061 additions and 2250 deletions
@@ -3,6 +3,71 @@
 ## 对话原则
 始终使用中文进行对话。

+## 编码规范 ⚠️ 重要
+
+### 文件编码要求
+- **所有文本文件必须使用 UTF-8 编码（无 BOM）**
+- **行结束符使用 LF（Unix 风格），Windows 批处理文件除外（CRLF）**
+- 项目已配置 `.gitattributes` 和 `.editorconfig` 来自动处理编码
+
+### 编辑器配置
+确保编辑器设置：
+- 文件编码：UTF-8（无 BOM）
+- 行结束符：LF
+- 自动插入文件末尾空行：开启
+
+## Git 提交规范
+
+### Commit Message 规范
+- **所有 commit message 必须使用英文**，避免中文编码问题
+- 使用 Conventional Commits 格式：`<type>: <description>`
+
+### 常用类型
+- `feat`: 新增功能
+- `fix`: 修复问题
+- `docs`: 文档更新
+- `refactor`: 代码重构
+- `style`: 代码格式调整
+- `test`: 测试相关
+- `chore`: 构建/工具配置
+
+### 示例
+```bash
+git commit -m "feat: add new model support"
+git commit -m "fix: resolve image encoding issue"
+git commit -m "docs: update README installation guide"
+```
+
+## 配置文件管理
+
+### 基本原则
+
+`.config` 文件包含敏感信息（API 密钥），已添加到 `.gitignore` 中，**不会被提交到版本控制**。
+
+### 配置方式
+
+用户通过以下方式创建本地配置：
+
+1. **快捷脚本**（推荐）
+   - Windows: 双击 `设置API密钥(win).bat`
+   - Linux/Mac: 运行 `./设置API密钥(mac).sh`
+   - 脚本会自动创建 `.config` 文件
+
+2. **手动创建**
+   - 参考 `.config.example` 模板
+   - 在插件根目录创建 `.config` 文件
+   - 填写 API 密钥
+
+3. **环境变量**
+   - 设置 `O1KEY_API_KEY` 环境变量
+   - 无需创建配置文件
+
+### 注意事项
+
+- `.config` 文件仅存在于本地，不会被 Git 追踪
+- 开发者无需担心意外提交密钥的问题
+- 提交代码时会自动忽略 `.config` 文件
+
 ## 项目概述

 这是一个 ComfyUI 自定义节点插件，通过 api.o1key.com 调用 AI 模型进行图像生成。
@@ -38,10 +103,13 @@ Comfyui_o1key/
 │   ├── __init__.py
 │   ├── base_client.py       # 客户端基类
 │   └── gemini_client.py     # Gemini API 客户端
-├── .config                  # API 配置文件（不提交）
-├── .config.example          # 配置示例
+├── .config.example          # 配置文件模板
 ├── requirements.txt         # 依赖包
 └── README.md                # 用户文档
+├── 设置API密钥(win).bat     # Windows 配置脚本
+└── 设置API密钥(mac).sh      # Mac/Linux 配置脚本
+
+注：.config 文件在本地自动创建，不提交到版本控制
 ```

 ---
@@ -366,7 +434,7 @@ pil_image = decode_base64_to_pil(b64_str)
 ### 配置管理 (utils/config.py)

 ```python
-from ..utils.config import get_api_key, get_api_key_or_raise, load_config
+from ..utils.config import get_api_key, get_api_key_or_raise, load_config, get_api_base_url

 # 获取 API 密钥（返回 None 如果未找到）
 api_key = get_api_key("O1KEY_API_KEY")
@@ -374,10 +442,64 @@ api_key = get_api_key("O1KEY_API_KEY")
 # 获取 API 密钥（抛出异常如果未找到）
 api_key = get_api_key_or_raise("O1KEY_API_KEY")

+# 获取 API 基础 URL（统一配置）
+base_url = get_api_base_url()  # 默认: https://vip.o1key.com
+
 # 加载完整配置
 config = load_config()
 ```

+### API 基础 URL 配置
+
+所有 API 客户端都使用统一的基础 URL 配置，默认为 `https://vip.o1key.com`。
+
+#### 配置优先级
+
+1. **环境变量** `O1KEY_API_BASE_URL`（优先级最高）
+2. **.config 文件**中的 `O1KEY_API_BASE_URL` 配置项
+3. **默认值** `https://vip.o1key.com`（在 `utils/config.py` 中定义）
+
+#### 修改 API 地址
+
+**方法 1：修改默认值（影响所有用户）**
+
+编辑 `utils/config.py`：
+
+```python
+# 修改此常量
+DEFAULT_API_BASE_URL = "https://your-api-domain.com"
+```
+
+**方法 2：使用环境变量（推荐，不影响代码）**
+
+在系统环境变量中设置：
+```bash
+# Windows
+set O1KEY_API_BASE_URL=https://your-api-domain.com
+
+# Linux/Mac
+export O1KEY_API_BASE_URL=https://your-api-domain.com
+```
+
+**方法 3：在 .config 文件中配置**
+
+在插件根目录的 `.config` 文件中添加：
+```
+O1KEY_API_BASE_URL=https://your-api-domain.com
+```
+
+#### 使用示例
+
+所有客户端会自动使用统一配置：
+
+```python
+from ..utils.config import get_api_base_url
+
+# 获取当前配置的 API 地址
+base_url = get_api_base_url()
+print(f"当前 API 地址: {base_url}")
+```
+
 ---

 ## API 客户端使用
@@ -0,0 +1,38 @@
+# EditorConfig 配置文件
+# https://editorconfig.org
+
+root = true
+
+# 默认配置
+[*]
+charset = utf-8
+end_of_line = lf
+insert_final_newline = true
+trim_trailing_whitespace = true
+indent_style = space
+indent_size = 4
+
+# Python 文件
+[*.py]
+indent_size = 4
+
+# Shell 脚本
+[*.sh]
+indent_size = 4
+
+# Windows 批处理文件
+[*.{bat,cmd}]
+end_of_line = crlf
+indent_size = 4
+
+# Markdown 文件
+[*.md]
+trim_trailing_whitespace = false
+
+# YAML 文件
+[*.{yml,yaml}]
+indent_size = 2
+
+# JSON 文件
+[*.json]
+indent_size = 2
@@ -0,0 +1,31 @@
+# 默认自动处理行结束符
+* text=auto
+
+# Python 文件使用 LF
+*.py text eol=lf
+
+# Shell 脚本使用 LF
+*.sh text eol=lf
+
+# Windows 批处理文件使用 CRLF
+*.bat text eol=crlf
+*.cmd text eol=crlf
+
+# 配置文件使用 LF
+.config text eol=lf
+.config.* text eol=lf
+
+# Markdown 文档使用 LF
+*.md text eol=lf
+
+# 二进制文件
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.gif binary
+*.ico binary
+*.mov binary
+*.mp4 binary
+*.mp3 binary
+*.zip binary
+*.psd binary
@@ -1,6 +1,3 @@
-# 隐私文件（已弃用配置文件，改用环境变量）
-# .config
-
 # Python 缓存
 __pycache__/
 *.py[cod]
@@ -24,3 +21,6 @@ venv/
 # OS
 .DS_Store
 Thumbs.db
+
+# 用户配置（含 API Key，不提交）
+.config
@@ -6,46 +6,59 @@

 ---

-## [1.10.0] - 2026-02-06
+## [1.10.4] - 2026-04-13

-### Added ⭐
- **自动更新系统** - 让用户轻松更新插件到最新版本
-  - 新增 `update.bat` - Windows 自动更新脚本
-  - 新增 `update.sh` - Linux/Mac 自动更新脚本
-  - 新增 `version.txt` - 版本号管理文件
-  - 新增 `utils/update_checker.py` - 启动时自动检查更新
-  - 新增更新检查功能：每次启动 ComfyUI 时自动检测是否有新版本
-  
- **更新脚本功能**：
-  - ✅ 自动检查远程更新
-  - ✅ 自动备份和恢复 `.config` 配置文件
-  - ✅ 自动拉取最新代码
-  - ✅ 自动更新 Python 依赖包
-  - ✅ 显示版本变更信息
-  - ✅ 显示最近更新日志（前 20 行）
-  - ✅ 友好的彩色终端输出（Linux/Mac）
-  - ✅ 完善的错误处理和提示
+### 修复
+- 修复香蕉2画草图导致生成2张图片问题
+
+---
+
+## [1.10.3] - 2026-04-13
+
+### 修复
+- 香蕉节点修复返回2张图、灰色图片问题
+- 修复 SSL 证书报错，统一由【保存图像】节点保存
+
+### 新增
+- 支持【立刻取消】生图请求，可立即重新运行
+- 支持工作流历史记录恢复
+- 并发上限提升，批量生图速度大幅提升
+- 全能LLM新增视频/文档分析、流式实时预览
+- Seedance 2.0 新增多图及视频URL参考（优化中）
+
+---
+
+## [Unreleased]
+
+### Added ✨
+- **快捷配置脚本**
+  - 新增 `设置API密钥(win).bat` - Windows 一键配置工具
+  - 新增 `设置API密钥(mac).sh` - Mac/Linux 一键配置工具
+  - 自动生成 `.config` 配置文件
+  - 交互式提示引导用户输入 API 密钥
+  - 自动检测并提示覆盖已存在的配置文件
+  - 彩色输出和友好的用户提示信息
+- **配置模板文件**
+  - 新增 `.config.example` 作为配置文件示例

 ### Changed
- **插件启动流程** (`__init__.py`)
-  - 集成更新检查模块
-  - 启动时自动检查是否有新版本
-  - 如有更新，终端显示友好的更新提示
-  - 静默失败机制，不影响插件正常加载
+- **502 错误提示优化** (`clients/base_client.py`)
+  - 当 API 返回 502 时，弹框显示友好文案：「糟糕！请求到上游时遇到超时或过载！别担心，过会儿再次点击运行即可！」
+  - 在 `request_async` 与 `request_get_async` 中均增加 502 专用分支
+- **配置管理策略**
+  - `.config` 文件现在完全忽略提交（添加到 `.gitignore`）
+  - 简化配置流程，用户通过快捷脚本自动创建本地配置
+  - 移除配置文件安全检查机制（不再需要）
+- **README 文档**
+  - 更新配置章节，添加快捷脚本使用说明
+  - 调整配置方法优先级：快捷脚本 > 环境变量 > 手动配置
+  - 简化安全提示说明

- **文档更新** (`README.md`)
-  - 新增"🔄 更新插件"章节
-  - 提供两种更新方法：自动更新（推荐）和手动更新
-  - 详细的跨平台更新说明
-  - 更新提示和注意事项
-
-### Benefits
- 🎯 **用户友好** - 一键更新，无需手动操作 Git
- 🔒 **配置安全** - 自动备份恢复配置，不会丢失设置
- ⚡ **依赖同步** - 自动更新 Python 包，确保兼容性
- 📋 **信息透明** - 显示版本变更和更新日志
- 🌍 **跨平台** - 支持 Windows/Linux/Mac
- 🛡️ **稳定可靠** - 完善的错误处理，不影响插件运行
+### Removed
+- **安全检查工具**（不再需要）
+  - 删除 `check_config_safety.py` 配置安全检查脚本
+  - 删除 `.git-hooks-install.bat` Git Hook 安装脚本
+  - 彻底杜绝配置文件泄密风险

 ---

@@ -10,3 +10,241 @@
 - 🎯 3 种分辨率（1K / 2K / 4K）
 - 🌱 可控随机种子

+---
+
+## 📦 安装
+
+### 方法一：通过 ComfyUI Manager（推荐）
+
+1. 在 ComfyUI 中打开 Manager
+2. 搜索 `Comfyui_o1key`
+3. 点击安装
+4. 重启 ComfyUI
+
+### 方法二：手动安装
+
+```bash
+cd ComfyUI/custom_nodes
+git clone https://github.com/lizhongyi1209/comfyui_o1key.git
+cd comfyui_o1key
+pip install -r requirements.txt
+```
+
+然后重启 ComfyUI。
+
+### 国内用户安装（GitHub 拉取慢或失败时）
+
+使用 Gitee 镜像安装与更新，避免网络问题：
+
+```bash
+cd ComfyUI/custom_nodes
+git clone https://gitee.com/resonLzy/comfyui_o1key.git
+cd comfyui_o1key
+pip install -r requirements.txt
+```
+
+自动更新脚本（见下方「更新插件」）已改为从 Gitee 拉取，国内用户可直接使用。
+
+---
+
+## ⚙️ 配置
+
+### 获取 API 密钥
+
+1. 访问 [vip.o1key.com](https://vip.o1key.com)
+2. 注册并获取 API 密钥
+
+### 配置方式
+
+#### 配置 API 密钥（必需）
+
+**方法一：快捷脚本配置（最简单）⭐**
+
+我们提供了一键配置脚本，自动创建配置文件：
+
+**Windows 用户：**
+双击运行 `设置API密钥(win).bat`，按提示输入 API 密钥即可。
+
+**Linux/Mac 用户：**
+```bash
+# 添加执行权限（仅首次需要）
+chmod +x 设置API密钥(mac).sh
+
+# 运行配置脚本
+./设置API密钥(mac).sh
+```
+
+按提示输入 API 密钥，配置完成后重启 ComfyUI。
+
+**方法二：环境变量（推荐）**
+
+**Windows 用户：**
+1. 右键 "此电脑" → 属性 → 高级系统设置 → 环境变量
+2. 在"用户变量"中新建：
+   - 变量名：`O1KEY_API_KEY`
+   - 变量值：你的 API 密钥
+3. 重启 ComfyUI
+
+**Linux/Mac 用户：**
+
+在 `~/.bashrc` 或 `~/.zshrc` 中添加：
+```bash
+export O1KEY_API_KEY="你的API密钥"
+```
+
+然后执行 `source ~/.bashrc` 并重启 ComfyUI。
+
+**方法三：手动创建配置文件**
+
+在插件目录下创建 `.config` 文件（参考 `.config.example`）：
+```
+O1KEY_API_KEY=你的API密钥
+```
+
+> **⚠️ 安全提示**
+> 
+> `.config` 文件包含敏感信息，已添加到 `.gitignore` 中，不会被提交到版本控制。
+> 请妥善保管你的 API 密钥，不要分享给他人。
+
+#### 配置 API 地址（可选）
+
+默认使用 `https://vip.o1key.com`，通常无需修改。
+
+如需自定义 API 地址，可通过以下方式：
+
+1. **环境变量**（推荐）：
+   ```bash
+   # Windows
+   set O1KEY_API_BASE_URL=https://your-api-domain.com
+   
+   # Linux/Mac
+   export O1KEY_API_BASE_URL=https://your-api-domain.com
+   ```
+
+2. **配置文件**：在 `.config` 中添加：
+   ```
+   O1KEY_API_BASE_URL=https://your-api-domain.com
+   ```
+
+3. **修改默认值**：编辑 `utils/config.py` 中的 `DEFAULT_API_BASE_URL` 常量
+
+---
+
+## 🔄 更新插件
+
+### 界面更新
+
+在 ComfyUI 左侧功能栏点击「更新」（位于「重启」下方）。按钮会从当前 Git 仓库的 `origin/main` 拉取最新版本。完成后点击「重启」使新版本生效。
+
+界面更新需要通过 Git 安装、处于 `main` 分支，且节点包文件没有本地修改。更新仅允许快进，不会覆盖本地修改或删除配置。ZIP 安装、分支分叉或网络连接失败时，界面会显示原因，需要手动处理。
+
+如果提示依赖列表已变化，请在 ComfyUI 使用的 Python 环境中执行：
+
+```bash
+cd ComfyUI/custom_nodes/comfyui_o1key
+python -m pip install -r requirements.txt
+```
+
+### 手动更新
+
+```bash
+cd ComfyUI/custom_nodes/comfyui_o1key
+git pull --ff-only origin main
+python -m pip install -r requirements.txt
+```
+
+更新保留环境变量中配置的 API 密钥。启动时仍会检查是否有新版本。
+
+---
+
+## 📚 节点说明
+
+### Nano Banana Pro
+
+高性能图像生成节点，支持文生图和图生图。
+
+**参数：**
+- **提示词**：描述你想生成的图像
+- **模型**：选择使用的 AI 模型
+- **分辨率**：1K / 2K / 4K
+- **宽高比**：1:1, 16:9, 9:16, 4:3, 3:4, 21:9, 9:21, 3:2, 2:3, 16:10
+- **批次大小**：单次生成的图像数量（1-1000）
+- **随机种子**：控制生成的随机性（-1 为随机）
+- **输入图像**（可选）：用于图生图模式
+
+### Batch Nano Banana Pro
+
+批量并发生成节点，适合大量图像生成。
+
+### Google Gemini
+
+Google Gemini 模型节点，支持更多模型选择。
+
+---
+
+## 📝 更新日志
+
+查看 [CHANGELOG.md](./CHANGELOG.md) 了解详细的版本更新记录。
+
+---
+
+## 📄 许可证
+
+本项目采用 Apache License 2.0 许可证。
+
+---
+
+## 🤝 贡献
+
+欢迎提交 Issue 和 Pull Request！
+
+---
+
+## ⚠️ 开发者注意事项
+
+### 维护者：发布流程与镜像同步
+
+代码**先提交并推送到 GitHub**，再**同步到 Gitee 镜像**，国内用户通过 Gitee 拉取以解决网络问题。
+
+**首次配置**（仅需一次）：
+```bash
+git remote add gitee https://gitee.com/resonLzy/comfyui_o1key.git
+```
+
+**每次发布**：
+```bash
+git push origin main    # 先更新 GitHub
+git push gitee main     # 再同步到 Gitee 镜像
+```
+
+### 文件编码要求
+
+**所有文本文件必须使用 UTF-8 编码（无 BOM）！**
+
+如果你在 GitHub 上看到中文乱码，说明文件编码有问题。请使用以下方法修复：
+
+**Windows 用户：**
+
+```powershell
+.\fix_encoding.ps1
+```
+
+**Linux/Mac 用户：**
+
+```bash
+chmod +x fix_encoding.sh
+./fix_encoding.sh
+```
+
+详细说明请查看 [编码修复指南.md](./编码修复指南.md)
+
+---
+
+## 📮 联系方式
+
+- GitHub: [@lizhongyi1209](https://github.com/lizhongyi1209)
+- 项目地址: https://github.com/lizhongyi1209/comfyui_o1key
+
+---
+
+**当前版本：v1.10.1**
@@ -0,0 +1,148 @@
+"""
+ComfyUI V3 节点开发参考
+========================
+
+本文件记录了将 V1 节点迁移到 V3 的关键经验，供后续节点开发快速参考。
+基于 nano_banana.py 的实际迁移总结。
+
+核心发现：V3 节点可以直接放入 V1 的 NODE_CLASS_MAPPINGS 中注册，
+ComfyUI 通过 issubclass(obj_class, _ComfyNodeInternal) 自动识别并
+调用 GET_NODE_INFO_V1() 生成前端所需的节点信息。无需 comfy_entrypoint。
+
+=== 最小 V3 节点模板 ===
+
+    from comfy_api.latest import io
+
+    class MyNode(io.ComfyNode):
+
+        @classmethod
+        def define_schema(cls):
+            return io.Schema(
+                node_id="MyNode",           # 必须与 NODE_CLASS_MAPPINGS 的 key 一致
+                display_name="我的节点",
+                category="image/generation",
+                inputs=[...],
+                outputs=[io.Image.Output(display_name="输出")],
+            )
+
+        @classmethod
+        def execute(cls, input1, input2, ...) -> io.NodeOutput:
+            # 业务逻辑
+            return io.NodeOutput(result)
+
+=== V1 → V3 对照表 ===
+
+    V1                              V3
+    ─────────────────────────────────────────────────────
+    INPUT_TYPES() classmethod       define_schema() → io.Schema
+    RETURN_TYPES = ("IMAGE",)       outputs=[io.Image.Output()]
+    RETURN_NAMES = ("输出",)        io.Image.Output(display_name="输出")
+    FUNCTION = "generate"           固定为 execute
+    CATEGORY = "xxx"                Schema(category="xxx")
+    generate(self, ...)             execute(cls, ...) classmethod
+    self.xxx 实例状态               模块级单例函数
+
+=== DynamicCombo（动态联动下拉框）===
+
+场景：一个 combo 的选项决定其他 combo 显示哪些值。
+
+    io.DynamicCombo.Input("模型", options=[
+        io.DynamicCombo.Option("选项A", [
+            io.Combo.Input("子参数1", options=["x", "y"]),
+            io.Combo.Input("子参数2", options=["1K", "2K"]),
+        ]),
+        io.DynamicCombo.Option("选项B", [
+            io.Combo.Input("子参数1", options=["x", "y", "z", "w"]),
+            io.Combo.Input("子参数2", options=["512px", "1K", "2K", "4K"]),
+        ]),
+    ])
+
+execute 中接收为 dict：
+    def execute(cls, 模型, ...):
+        selected = 模型["模型"]      # "选项A" 或 "选项B"
+        sub1 = 模型["子参数1"]       # 对应选项下的子输入值
+        sub2 = 模型["子参数2"]
+
+注意：dict 的 key 是 DynamicCombo.Input 的 id（"模型"），
+子输入的 key 是各 Combo.Input 的 id。
+
+=== Autogrow（自动增长输入槽）===
+
+场景：用户连接一个槽后自动出现下一个，最多 N 个。
+
+    io.Autogrow.Input("参考图",
+        template=io.Autogrow.TemplatePrefix(
+            input=io.Image.Input("img"),
+            prefix="参考图",        # 生成 参考图0, 参考图1, ...
+            min=0,                  # 最少显示几个槽
+            max=9,                  # 最多几个槽
+        ),
+    )
+
+execute 中接收为 dict（或 io.Autogrow.Type）：
+    def execute(cls, 参考图=None, ...):
+        if 参考图:
+            for key, tensor in 参考图.items():
+                # key = "参考图0", "参考图1", ...
+                # tensor = IMAGE tensor 或 None
+
+=== 实例状态处理 ===
+
+V3 的 execute 是 classmethod，无法用 self。
+用模块级单例替代：
+
+    _client = None
+
+    def _get_client():
+        global _client
+        if _client is None:
+            _client = MyAPIClient()
+        return _client
+
+=== 注册方式（与 V1 共存）===
+
+在 __init__.py 中照常注册，无需任何特殊处理：
+
+    NODE_CLASS_MAPPINGS = {
+        "MyV1Node": MyV1Node,       # V1 节点
+        "MyV3Node": MyV3Node,       # V3 节点，自动识别
+    }
+
+    NODE_DISPLAY_NAME_MAPPINGS = {
+        "MyV1Node": "V1 节点",
+        "MyV3Node": "V3 节点",      # 也可省略，V3 用 Schema.display_name
+    }
+
+=== 注意事项 ===
+
+1. node_id 必须与 NODE_CLASS_MAPPINGS 的 key 完全一致
+2. V3 execute 返回 io.NodeOutput(tensor)，不是 tuple
+3. _wrap_generate_for_error_display 等 V1 包装器对 V3 无效
+   （找不到 generate 方法会安全跳过）
+4. V3 支持 async execute（直接加 async 即可）
+5. 输入参数名必须与 Schema inputs 的 id 一致
+6. DynamicCombo 的子输入在前端会随选项切换动态显示/隐藏
+7. Autogrow 的 widget 输入会被强制为 force_input（仅连接，无控件）
+
+=== 可用输入类型速查 ===
+
+    io.String.Input(id, default="", multiline=False)
+    io.Int.Input(id, default=0, min=0, max=N, step=1)
+    io.Float.Input(id, default=0.0, min=0.0, max=N, step=0.01)
+    io.Combo.Input(id, options=[...], default="...")
+    io.Boolean.Input(id, default=False)
+    io.Image.Input(id)
+    io.Mask.Input(id)
+    io.Latent.Input(id)
+    io.DynamicCombo.Input(id, options=[DynamicCombo.Option(...)])
+    io.Autogrow.Input(id, template=TemplatePrefix/TemplateNames)
+
+=== 可用输出类型速查 ===
+
+    io.Image.Output(display_name="...")
+    io.String.Output(display_name="...")
+    io.Int.Output()
+    io.Float.Output()
+    io.Latent.Output()
+    io.Mask.Output()
+"""
@@ -9,29 +9,747 @@ Comfyui_o1key - ComfyUI 自定义节点集合
 └── __init__.py     # 节点注册入口
 """

-# 检查更新（仅在启动时检查一次）
+
+import ssl
+import logging
+import asyncio
+
+# 屏蔽 ComfyUI 资产扫描的终端日志输出
+_seeder_filter = lambda record: not any(
+    kw in record.getMessage()
+    for kw in ("Seeder start", "Asset scan", "Scan(", "Fast scan")
+)
+logging.getLogger().addFilter(_seeder_filter)
+
+
+def _is_ignored_asyncio_win10054(context):
+    exc = context.get("exception")
+    if not (
+        isinstance(exc, ConnectionResetError)
+        and getattr(exc, "winerror", None) == 10054
+    ):
+        return False
+
+    handle = str(context.get("handle", ""))
+    message = str(context.get("message", ""))
+    marker = "_ProactorBasePipeTransport._call_connection_lost"
+    return marker in handle or marker in message
+
+
+def _install_asyncio_win10054_filter(loop):
+    if getattr(loop, "_o1key_win10054_filter_installed", False):
+        return loop
+
+    previous_handler = loop.get_exception_handler()
+
+    def _o1key_asyncio_exception_handler(loop, context):
+        if _is_ignored_asyncio_win10054(context):
+            return
+        if previous_handler is not None:
+            previous_handler(loop, context)
+        else:
+            loop.default_exception_handler(context)
+
+    loop.set_exception_handler(_o1key_asyncio_exception_handler)
+    setattr(loop, "_o1key_win10054_filter_installed", True)
+    return loop
+
+
 try:
-    from .utils.update_checker import check_for_updates, notify_update_available
-    
-    if check_for_updates():
-        notify_update_available()
-except Exception:
-    # 静默失败，不影响插件加载
+    _install_asyncio_win10054_filter(asyncio.get_event_loop())
+except RuntimeError:
    pass

-from .nodes import NanoBananaPro, BatchNanoBananaPro, GoogleGemini
+if not getattr(asyncio, "_o1key_new_event_loop_patched", False):
+    _o1key_original_new_event_loop = asyncio.new_event_loop
+
+    def _o1key_new_event_loop(*args, **kwargs):
+        return _install_asyncio_win10054_filter(
+            _o1key_original_new_event_loop(*args, **kwargs)
+        )
+
+    asyncio.new_event_loop = _o1key_new_event_loop
+    asyncio._o1key_new_event_loop_patched = True
+
+from .nodes import NanoBananaPro, BatchNanoBananaPro, GoogleGemini, LoadFile, ImageStitchPro, BatchCleanMetadata, VideoPreview, GoogleVeo, Google31Video, FluxImageEdit, UniversalLLMChat, KlingVideo, KlingFirstLastFrame, KlingMotionControlTest, AspectRatioPreset, BatchImagesO1key, Seedance, SeedanceMultiModal, StreamPreview, DoubaoImage, O1keyGPTImage, O1keyGPTImageBatch, O1keyGrokImage, O1keyGrokVideo, KVideoFirstLast, KVideoImage2Video
+from .nodes import K3Video, K3VideoFirstLast, K3MotionControl, K3MotionVideoCheck, NanoBananaV2, NanoBananaV2Batch, SaveImageFormat
+from .nodes import O1keySavePSD
+from .nodes import O1keyRemoveBackground
+from .nodes import O1keyColorRemoveBG
+from .nodes import O1keyGridSplitter
+
+# 报错弹框友好文案（不修改原节点代码，仅在外层统一处理）
+_MSG_TIMEOUT = "API 请求超时，请稍后重试或检查网络。"
+_MSG_SSL_NETWORK = (
+    "本地网络不太稳定!解决方案如下:\n"
+    "1. 重启程序再试试看 (优先)\n"
+    "2. 调整一下网络环境,如wifi或宽带等\n"
+    "3. 切换VPN节点,或更换代理模式\n"
+    "4. 关掉杀毒软件或防火墙\n"
+    "5. 关掉浏览器VPN插件,避免冲突"
+)
+
+def _wrap_generate_for_error_display(cls, attr="generate"):
+    original = getattr(cls, attr, None)
+    if original is None:
+        return
+    def wrapped(self, *args, **kwargs):
+        try:
+            return original(self, *args, **kwargs)
+        except TimeoutError as e:
+            msg = (str(e) or "").strip()
+            if not msg:
+                msg = _MSG_TIMEOUT
+            raise TimeoutError(msg) from None
+        except (ssl.SSLError, OSError) as e:
+            err_str = str(e)
+            if "DECRYPTION_FAILED_OR_BAD_RECORD_MAC" in err_str or "decryption failed or bad record mac" in err_str.lower():
+                raise RuntimeError(_MSG_SSL_NETWORK) from None
+            raise
+    setattr(cls, attr, wrapped)
+
+_wrap_generate_for_error_display(NanoBananaPro)
+_wrap_generate_for_error_display(BatchNanoBananaPro)
+_wrap_generate_for_error_display(NanoBananaV2)
+_wrap_generate_for_error_display(NanoBananaV2Batch)

 # ComfyUI 节点注册
 NODE_CLASS_MAPPINGS = {
-    "NanoBananaPro": NanoBananaPro,
+    "NanoBanana": NanoBananaPro,
    "BatchNanoBananaPro": BatchNanoBananaPro,
-    "GoogleGemini": GoogleGemini
+    "GoogleGemini": GoogleGemini,
+    "LoadFile": LoadFile,
+    "ImageStitchPro": ImageStitchPro,
+
+    "BatchCleanMetadata": BatchCleanMetadata,
+    "VideoPreview": VideoPreview,
+    "GoogleVeo": GoogleVeo,
+    "Google31Video": Google31Video,
+    "FluxImageEdit": FluxImageEdit,
+    "UniversalLLMChat": UniversalLLMChat,
+    "KlingVideo": KlingVideo,
+    "KlingFirstLastFrame": KlingFirstLastFrame,
+    "KlingMotionControlTest": KlingMotionControlTest,
+    "AspectRatioPreset": AspectRatioPreset,
+
+    "BatchImagesO1key": BatchImagesO1key,
+    "Seedance": Seedance,
+    "SeedanceMultiModal": SeedanceMultiModal,
+    "StreamPreview": StreamPreview,
+    "DoubaoImage": DoubaoImage,
+    "O1keyGPTImage": O1keyGPTImage,
+    "O1keyGPTImageBatch": O1keyGPTImageBatch,
+    "O1keyGrokImage": O1keyGrokImage,
+    "O1keyGrokVideo": O1keyGrokVideo,
+    "KVideoFirstLast": KVideoFirstLast,
+    "KVideoImage2Video": KVideoImage2Video,
+    "K3Video": K3Video,
+    "K3VideoFirstLast": K3VideoFirstLast,
+    "K3MotionControl": K3MotionControl,
+    "K3MotionVideoCheck": K3MotionVideoCheck,
+    "NanoBananaV2": NanoBananaV2,
+    "NanoBananaV2Batch": NanoBananaV2Batch,
+    "SaveImageFormat": SaveImageFormat,
+    "O1keySavePSD": O1keySavePSD,
+    "O1keyRemoveBackground": O1keyRemoveBackground,
+    "O1keyColorRemoveBG": O1keyColorRemoveBG,
+    "O1keyGridSplitter": O1keyGridSplitter,
 }

 NODE_DISPLAY_NAME_MAPPINGS = {
-    "NanoBananaPro": "Nano Banana Pro",
-    "BatchNanoBananaPro": "批量 Nano Banana Pro",
-    "GoogleGemini": "Google Gemini"
+    "NanoBanana": "Nano Banana",
+    "BatchNanoBananaPro": "批量 Nano Banana",
+    "GoogleGemini": "Google Gemini",
+    "LoadFile": "加载文件",
+    "ImageStitchPro": "图像拼接 Pro",
+
+    "BatchCleanMetadata": "批量任务（防AI识别）",
+    "VideoPreview": "预览视频",
+    "GoogleVeo": "Google Veo - ab",
+    "Google31Video": "Google 3.1 Video",
+    "FluxImageEdit": "Flux2 图像编辑",
+    "UniversalLLMChat": "全能LLM对话助手",
+    "KlingVideo": "文/图生视频 自研模型",
+    "KlingFirstLastFrame": "首尾帧生视频 自研模型",
+    "KlingMotionControlTest": "动作控制 自研模型",
+    "AspectRatioPreset": "图片宽高比预设",
+
+    "BatchImagesO1key": "加载图像（批量）",
+    "Seedance": "Seedance 视频生成",
+    "SeedanceMultiModal": "Seedance 多模态参考生视频",
+    "StreamPreview": "流式文本预览",
+    "DoubaoImage": "豆包生图",
+    "O1keyGPTImage": "o1key GPT Image",
+    "O1keyGPTImageBatch": "o1key GPT Image（批量）",
+    "O1keyGrokImage": "Grok Image",
+    "O1keyGrokVideo": "Grok Video",
+    "KVideoFirstLast": "K26 图生视频（首尾帧）",
+    "KVideoImage2Video": "K26 图生视频",
+    "K3Video": "K3 图生视频 自研",
+    "K3VideoFirstLast": "首尾帧 K3 自研",
+    "K3MotionControl": "动作控制 K3 自研",
+    "K3MotionVideoCheck": "视频时长检测 K3",
+    "NanoBananaV2": "Nano Banana V2",
+    "NanoBananaV2Batch": "Nano Banana V2（批量）",
+    "SaveImageFormat": "保存图像（格式转换）",
+    "O1keySavePSD": "保存 PSD（分层）",
+    "O1keyRemoveBackground": "去背景（rembg）",
+    "O1keyColorRemoveBG": "颜色去背景",
+    "O1keyGridSplitter": "合并图智能切割",
 }

-__all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS']
+WEB_DIRECTORY = "./web"
+
+__all__ = ['NODE_CLASS_MAPPINGS', 'NODE_DISPLAY_NAME_MAPPINGS', 'WEB_DIRECTORY']
+
+# 注册 /o1key/input_dir 接口，供前端文件上传按钮获取 input 目录绝对路径
+try:
+    from aiohttp import web
+    from server import PromptServer
+    import folder_paths
+    from .utils.config import CONFIG_FILE, load_config, NETWORK_ROUTES
+    from .utils.updater import UpdateError, update_package
+    import threading as _update_threading
+
+    _update_lock = _update_threading.Lock()
+
+    def _get_o1key_server_port():
+        try:
+            import comfy.cli_args as _cli_args
+            args = getattr(_cli_args, "args", None)
+            port = getattr(args, "port", None) if args else None
+            port = port or getattr(_cli_args, "server_port", None) or getattr(_cli_args, "port", None)
+            if port is not None:
+                return str(int(port))
+        except Exception:
+            pass
+        try:
+            import sys as _sys
+            for idx, arg in enumerate(_sys.argv):
+                if arg in ("--port", "--listen-port") and idx + 1 < len(_sys.argv):
+                    return str(int(_sys.argv[idx + 1]))
+                for prefix in ("--port=", "--listen-port="):
+                    if arg.startswith(prefix):
+                        return str(int(arg.split("=", 1)[1]))
+        except Exception:
+            pass
+        return "8188"
+
+    def _get_o1key_history_meta_file(output_dir):
+        import os as _os_history
+        return _os_history.path.join(
+            output_dir,
+            f".o1key_history_{_get_o1key_server_port()}.json",
+        )
+
+    def _get_o1key_notes_file():
+        import os as _os_notes
+        input_dir = _os_notes.path.abspath(folder_paths.get_input_directory())
+        _os_notes.makedirs(input_dir, exist_ok=True)
+        return _os_notes.path.join(input_dir, "o1key-notes.json")
+
+    def _extract_o1key_notes(payload):
+        if isinstance(payload, list):
+            return payload
+        if isinstance(payload, dict) and isinstance(payload.get("notes"), list):
+            return payload["notes"]
+        return None
+
+    @PromptServer.instance.routes.get("/o1key/notes")
+    async def get_o1key_notes(request):
+        import os as _os_notes
+        import json as _json_notes
+
+        notes_file = _get_o1key_notes_file()
+        exists = _os_notes.path.isfile(notes_file)
+        notes = []
+
+        if exists:
+            try:
+                with open(notes_file, "r", encoding="utf-8") as nf:
+                    loaded = _json_notes.load(nf)
+                notes = _extract_o1key_notes(loaded)
+                if notes is None:
+                    return web.json_response(
+                        {"error": "invalid notes file", "path": notes_file},
+                        status=500,
+                    )
+            except Exception as e:
+                return web.json_response(
+                    {"error": str(e), "path": notes_file},
+                    status=500,
+                )
+
+        return web.json_response({"notes": notes, "path": notes_file, "exists": exists})
+
+    @PromptServer.instance.routes.post("/o1key/notes")
+    async def save_o1key_notes(request):
+        import os as _os_notes
+        import json as _json_notes
+
+        try:
+            payload = await request.json()
+            notes = _extract_o1key_notes(payload)
+            if notes is None:
+                return web.json_response({"error": "notes must be a list"}, status=400)
+        except Exception as e:
+            return web.json_response({"error": f"invalid notes payload: {str(e)}"}, status=400)
+
+        notes_file = _get_o1key_notes_file()
+        temp_file = notes_file + ".tmp"
+        try:
+            with open(temp_file, "w", encoding="utf-8") as nf:
+                _json_notes.dump(notes, nf, ensure_ascii=False, indent=2)
+                nf.write("\n")
+            _os_notes.replace(temp_file, notes_file)
+        except Exception as e:
+            return web.json_response({"error": f"save notes failed: {str(e)}"}, status=500)
+
+        return web.json_response({
+            "success": True,
+            "path": notes_file,
+            "count": len(notes),
+        })
+
+    @PromptServer.instance.routes.get("/o1key/input_dir")
+    async def get_input_dir(request):
+        import os
+        path = os.path.abspath(folder_paths.get_input_directory())
+        return web.json_response({"path": path})
+
+    @PromptServer.instance.routes.get("/o1key/api_key")
+    async def get_api_key_route(request):
+        config = load_config()
+        key = config.get("O1KEY_API_KEY", "")
+        masked = ""
+        if key:
+            if len(key) > 8:
+                masked = key[:3] + "****" + key[-4:]
+            else:
+                masked = "****"
+        return web.json_response({"has_key": bool(key), "masked": masked})
+
+    @PromptServer.instance.routes.post("/o1key/api_key")
+    async def set_api_key_route(request):
+        import os
+        data = await request.json()
+        new_key = data.get("api_key", "").strip()
+        if not new_key:
+            return web.json_response({"error": "API Key 不能为空"}, status=400)
+        config = load_config()
+        config["O1KEY_API_KEY"] = new_key
+        lines = []
+        for k, v in config.items():
+            lines.append(f"{k}={v}")
+        with open(CONFIG_FILE, 'w', encoding='utf-8') as f:
+            f.write("\n".join(lines) + "\n")
+        return web.json_response({"success": True})
+
+    @PromptServer.instance.routes.post("/o1key/test_key")
+    async def test_api_key_route(request):
+        import aiohttp as _aiohttp
+        data = await request.json()
+        test_key = data.get("api_key", "").strip()
+        if not test_key:
+            return web.json_response({"valid": False, "error": "密钥不能为空"})
+        base_url = NETWORK_ROUTES.get("CF加速", "https://cf-api.o1key.com")
+        url = f"{base_url}/v1/models"
+        headers = {"Authorization": f"Bearer {test_key}"}
+        try:
+            async with _aiohttp.ClientSession() as session:
+                async with session.get(url, headers=headers, timeout=_aiohttp.ClientTimeout(total=10)) as resp:
+                    if resp.status == 200:
+                        return web.json_response({"valid": True})
+                    elif resp.status == 401:
+                        return web.json_response({"valid": False, "error": "密钥无效或已过期"})
+                    else:
+                        text = await resp.text()
+                        return web.json_response({"valid": False, "error": f"验证失败 ({resp.status})"})
+        except Exception as e:
+            return web.json_response({"valid": False, "error": f"网络错误: {str(e)}"})
+
+    @PromptServer.instance.routes.delete("/o1key/api_key")
+    async def delete_api_key_route(request):
+        config = load_config()
+        config.pop("O1KEY_API_KEY", None)
+        lines = []
+        for k, v in config.items():
+            lines.append(f"{k}={v}")
+        with open(CONFIG_FILE, 'w', encoding='utf-8') as f:
+            f.write("\n".join(lines) + "\n")
+        return web.json_response({"success": True})
+
+    @PromptServer.instance.routes.get("/o1key/output_history")
+    async def get_output_history(request):
+        """读取 output 目录文件，按执行分组返回 /api/jobs 兼容格式"""
+        import os, json as _json
+        limit = int(request.query.get("limit", "200"))
+        offset = int(request.query.get("offset", "0"))
+        output_dir = os.path.abspath(folder_paths.get_output_directory())
+        meta_file = _get_o1key_history_meta_file(output_dir)
+        meta = {}
+        if os.path.isfile(meta_file):
+            try:
+                with open(meta_file, "r", encoding="utf-8") as mf:
+                    meta = _json.load(mf)
+            except Exception:
+                pass
+        supported_ext = {'.png', '.jpg', '.jpeg', '.webp', '.gif', '.mp4', '.webm'}
+        # 收集所有文件并按 workflow_id 分组
+        all_files = []
+        for fname in meta.keys():
+            ext = os.path.splitext(fname)[1].lower()
+            if ext not in supported_ext:
+                continue
+            fpath = os.path.join(output_dir, fname)
+            if not os.path.isfile(fpath):
+                continue
+            mtime = os.path.getmtime(fpath)
+            media = "images" if ext in {'.png','.jpg','.jpeg','.webp','.gif'} else "video"
+            all_files.append({"name": fname, "mtime": mtime, "media": media})
+        # 按 workflow_id 分组（同一次执行合并为一个 job）
+        groups = {}
+        for f in all_files:
+            m = meta.get(f["name"], {})
+            wid = m.get("workflow_id")
+            if wid:
+                groups.setdefault(wid, []).append((f, m))
+        # 构建 job 列表
+        jobs = []
+        for wid, items in groups.items():
+            items.sort(key=lambda x: x[0]["mtime"], reverse=True)
+            latest = items[0]
+            f, m = latest
+            start_ms = int(m.get("start_time", f["mtime"]) * 1000)
+            end_ms = int(m.get("end_time", f["mtime"]) * 1000)
+            jobs.append({
+                "id": wid,
+                "status": "completed",
+                "create_time": start_ms,
+                "execution_start_time": start_ms,
+                "execution_end_time": end_ms,
+                "preview_output": {
+                    "filename": f["name"],
+                    "subfolder": "",
+                    "type": "output",
+                    "nodeId": "0",
+                    "mediaType": f["media"],
+                },
+                "outputs_count": len(items),
+                "execution_error": None,
+                "workflow_id": wid,
+            })
+        # 按时间倒序排列，分页
+        jobs.sort(key=lambda x: x["create_time"], reverse=True)
+        total = len(jobs)
+        page = jobs[offset:offset+limit]
+        return web.json_response({
+            "jobs": page,
+            "pagination": {"offset": offset, "limit": limit, "total": total, "has_more": offset + limit < total}
+        })
+
+    @PromptServer.instance.routes.get("/o1key/output_workflow")
+    async def get_output_workflow(request):
+        """从 PNG 元数据中读取工作流，供前端恢复使用"""
+        import os, struct, json as _json
+        filename = request.query.get("filename", "")
+        if not filename:
+            return web.json_response({"error": "missing filename"}, status=400)
+        output_dir = os.path.abspath(folder_paths.get_output_directory())
+        fpath = os.path.join(output_dir, filename)
+        if not os.path.isfile(fpath) or not fpath.lower().endswith(".png"):
+            return web.json_response({"error": "file not found"}, status=404)
+        workflow = None
+        prompt_data = None
+        try:
+            with open(fpath, "rb") as pf:
+                pf.read(8)  # PNG signature
+                while True:
+                    raw = pf.read(8)
+                    if len(raw) < 8:
+                        break
+                    length = struct.unpack(">I", raw[:4])[0]
+                    chunk_type = raw[4:8]
+                    data = pf.read(length)
+                    pf.read(4)  # CRC
+                    if chunk_type == b"tEXt":
+                        key, val = data.split(b"\x00", 1)
+                        k = key.decode("ascii", errors="replace")
+                        if k == "workflow":
+                            workflow = _json.loads(val)
+                        elif k == "prompt":
+                            prompt_data = _json.loads(val)
+                    elif chunk_type == b"IEND":
+                        break
+        except Exception:
+            pass
+        return web.json_response({"workflow": workflow, "prompt": prompt_data})
+
+    @PromptServer.instance.routes.get("/o1key/job_detail/{job_id}")
+    async def get_job_detail(request):
+        """根据 job_id 返回当前端口持久化历史中的 job 详情"""
+        import os, struct, json as _json
+        job_id = request.match_info["job_id"]
+        output_dir = os.path.abspath(folder_paths.get_output_directory())
+        meta_file = _get_o1key_history_meta_file(output_dir)
+        meta = {}
+        if os.path.isfile(meta_file):
+            try:
+                with open(meta_file, "r", encoding="utf-8") as mf:
+                    meta = _json.load(mf)
+            except Exception:
+                pass
+        supported_ext = {'.png', '.jpg', '.jpeg', '.webp', '.gif', '.mp4', '.webm'}
+        # 只在当前端口的持久化记录中查找该 job 的文件
+        matched_files = []
+        for fname in meta.keys():
+            ext = os.path.splitext(fname)[1].lower()
+            if ext not in supported_ext:
+                continue
+            m = meta.get(fname, {})
+            if m.get("workflow_id") == job_id:
+                matched_files.append(fname)
+        if not matched_files:
+            return web.json_response({"error": "not found"}, status=404)
+        # 用最新文件作为代表
+        matched_files.sort(key=lambda f: os.path.getmtime(os.path.join(output_dir, f)), reverse=True)
+        target_file = matched_files[0]
+        fpath = os.path.join(output_dir, target_file)
+        m = meta.get(target_file, {})
+        mtime = os.path.getmtime(fpath)
+        start_ms = int(m.get("start_time", mtime) * 1000)
+        end_ms = int(m.get("end_time", mtime) * 1000)
+        ext = os.path.splitext(target_file)[1].lower()
+        media = "images" if ext in {'.png','.jpg','.jpeg','.webp','.gif'} else "video"
+        # 读取 PNG 工作流元数据
+        workflow = None
+        if ext == ".png":
+            try:
+                with open(fpath, "rb") as pf:
+                    pf.read(8)
+                    while True:
+                        raw = pf.read(8)
+                        if len(raw) < 8:
+                            break
+                        length = struct.unpack(">I", raw[:4])[0]
+                        chunk_type = raw[4:8]
+                        data = pf.read(length)
+                        pf.read(4)
+                        if chunk_type == b"tEXt":
+                            key, val = data.split(b"\x00", 1)
+                            k = key.decode("ascii", errors="replace")
+                            if k == "workflow":
+                                workflow = _json.loads(val)
+                        elif chunk_type == b"IEND":
+                            break
+            except Exception:
+                pass
+        # 构建 outputs：包含该执行的所有文件
+        outputs = {}
+        for i, fname in enumerate(matched_files):
+            e = os.path.splitext(fname)[1].lower()
+            mt = "images" if e in {'.png','.jpg','.jpeg','.webp','.gif'} else "gifs"
+            outputs.setdefault(str(i), {}).setdefault(mt, []).append(
+                {"filename": fname, "subfolder": "", "type": "output"}
+            )
+        job_detail = {
+            "id": job_id,
+            "status": "completed",
+            "create_time": start_ms,
+            "execution_start_time": start_ms,
+            "execution_end_time": end_ms,
+            "preview_output": {
+                "filename": target_file,
+                "subfolder": "",
+                "type": "output",
+                "nodeId": "0",
+                "mediaType": media,
+            },
+            "outputs_count": len(matched_files),
+            "execution_error": None,
+            "workflow_id": job_id,
+            "workflow": {
+                "extra_data": {
+                    "extra_pnginfo": {"workflow": workflow}
+                }
+            } if workflow else None,
+            "outputs": outputs,
+        }
+        return web.json_response(job_detail)
+
+    @PromptServer.instance.routes.post("/o1key/delete_history")
+    async def delete_history_item(request):
+        """删除持久化历史记录及对应的输出文件"""
+        import os, json as _json
+        body = await request.json()
+        job_ids = body.get("delete", [])
+        if not job_ids:
+            return web.json_response({"success": False, "error": "missing ids"}, status=400)
+        output_dir = os.path.abspath(folder_paths.get_output_directory())
+        meta_file = _get_o1key_history_meta_file(output_dir)
+        meta = {}
+        if os.path.isfile(meta_file):
+            try:
+                with open(meta_file, "r", encoding="utf-8") as mf:
+                    meta = _json.load(mf)
+            except Exception:
+                pass
+        deleted_files = []
+        for job_id in job_ids:
+            files_to_remove = []
+            for fname, m in list(meta.items()):
+                if m.get("workflow_id") == job_id:
+                    files_to_remove.append(fname)
+            for fname in files_to_remove:
+                meta.pop(fname, None)
+                fpath = os.path.join(output_dir, fname)
+                if os.path.isfile(fpath):
+                    try:
+                        os.remove(fpath)
+                        deleted_files.append(fname)
+                    except Exception:
+                        pass
+        try:
+            with open(meta_file, "w", encoding="utf-8") as mf:
+                _json.dump(meta, mf, ensure_ascii=False)
+        except Exception:
+            pass
+        return web.json_response({"success": True, "deleted": deleted_files})
+
+    @PromptServer.instance.routes.post("/o1key/update")
+    async def update_node_package(request):
+        if request.headers.get("X-O1Key-Update") != "1":
+            return web.json_response({"error": "无效的更新请求。"}, status=403)
+        if not _update_lock.acquire(blocking=False):
+            return web.json_response({"error": "更新正在进行，请稍候。"}, status=409)
+        try:
+            result = await asyncio.to_thread(update_package)
+            return web.json_response(result)
+        except UpdateError as exc:
+            return web.json_response({"error": str(exc)}, status=409)
+        except Exception:
+            logging.exception("o1key update failed")
+            return web.json_response({"error": "更新失败，请查看 ComfyUI 日志。"}, status=500)
+        finally:
+            _update_lock.release()
+
+    # === AI 聊天代理（流式 SSE 透传） ===
+    @PromptServer.instance.routes.post("/o1key/restart")
+    async def restart_server(request):
+        import sys, os as _ros, subprocess, threading
+        def _do_restart():
+            import time
+            time.sleep(1.5)
+            skip = {"--auto-launch", "--auto_launch", "--launch", "--windows-standalone-build"}
+            args = [a for a in sys.argv if a not in skip]
+            args.append("--disable-auto-launch")
+            subprocess.Popen([sys.executable] + args, cwd=_ros.getcwd())
+            _ros._exit(0)
+        threading.Thread(target=_do_restart, daemon=True).start()
+        return web.json_response({"success": True, "message": "正在重启..."})
+
+    # === AI 聊天代理（流式 SSE 透传） ===
+    @PromptServer.instance.routes.post("/o1key/chat/completions")
+    async def chat_completions_proxy(request):
+        import aiohttp as _aiohttp
+        import json as _cjson
+
+        data = await request.json()
+        config = load_config()
+        api_key = config.get("O1KEY_API_KEY", "")
+        if not api_key:
+            return web.json_response({"error": "未配置 API Key"}, status=401)
+
+        route = data.get("route", "CF加速")
+        base_url = NETWORK_ROUTES.get(route, "https://cf-api.o1key.com")
+        model = data.get("model", "gpt-5.5")
+        messages = data.get("messages", [])
+
+        url = f"{base_url}/v1/chat/completions"
+        headers = {
+            "Content-Type": "application/json",
+            "Authorization": f"Bearer {api_key}",
+        }
+        body = {"model": model, "messages": messages, "stream": True}
+
+        resp = web.StreamResponse(
+            status=200, reason="OK",
+            headers={
+                "Content-Type": "text/event-stream",
+                "Cache-Control": "no-cache",
+                "X-Accel-Buffering": "no",
+            }
+        )
+        await resp.prepare(request)
+
+        try:
+            timeout = _aiohttp.ClientTimeout(total=120)
+            async with _aiohttp.ClientSession(timeout=timeout) as session:
+                async with session.post(url, headers=headers, json=body) as upstream:
+                    if upstream.status != 200:
+                        err = await upstream.text()
+                        await resp.write(f"data: {_cjson.dumps({'error': err})}\n\n".encode())
+                        await resp.write(b"data: [DONE]\n\n")
+                        return resp
+                    async for chunk in upstream.content.iter_any():
+                        await resp.write(chunk)
+        except Exception as e:
+            await resp.write(f"data: {_cjson.dumps({'error': str(e)})}\n\n".encode())
+            await resp.write(b"data: [DONE]\n\n")
+
+        return resp
+
+    # === 执行事件 Hook：持久化耗时元数据 ===
+    import time as _time, json as _json2, os as _os
+    _execution_tracker = {}
+
+    _orig_send_sync = PromptServer.instance.send_sync
+
+    def _patched_send_sync(event, data, *args, **kwargs):
+        try:
+            if event == "execution_start":
+                pid = data.get("prompt_id", "")
+                if pid:
+                    _execution_tracker[pid] = {"start": _time.time(), "outputs": []}
+            elif event == "executed":
+                pid = data.get("prompt_id", "")
+                output = data.get("output") or {}
+                if pid and pid in _execution_tracker:
+                    for img in output.get("images", []) + output.get("gifs", []):
+                        if img.get("type") == "output" and img.get("filename"):
+                            _execution_tracker[pid]["outputs"].append(img["filename"])
+            elif event == "executing" and data.get("node") is None:
+                pid = data.get("prompt_id", "")
+                tracker = _execution_tracker.pop(pid, None)
+                if tracker and tracker["outputs"]:
+                    end_time = _time.time()
+                    start_time = tracker["start"]
+                    output_dir = _os.path.abspath(folder_paths.get_output_directory())
+                    meta_file = _get_o1key_history_meta_file(output_dir)
+                    meta = {}
+                    if _os.path.isfile(meta_file):
+                        try:
+                            with open(meta_file, "r", encoding="utf-8") as mf:
+                                meta = _json2.load(mf)
+                        except Exception:
+                            pass
+                    for fname in tracker["outputs"]:
+                        meta[fname] = {
+                            "start_time": start_time,
+                            "end_time": end_time,
+                            "outputs_count": len(tracker["outputs"]),
+                            "workflow_id": pid,
+                        }
+                    try:
+                        with open(meta_file, "w", encoding="utf-8") as mf:
+                            _json2.dump(meta, mf, ensure_ascii=False)
+                    except Exception:
+                        pass
+        except Exception:
+            pass
+        return _orig_send_sync(event, data, *args, **kwargs)
+
+    PromptServer.instance.send_sync = _patched_send_sync
+
+except Exception:
+    pass
@@ -6,5 +6,11 @@ API 客户端模块
 from .base_client import BaseAPIClient
 from .gemini_client import GeminiAPIClient
 from .gemini_flash_client import GeminiFlashClient
+from .sora_client import SoraClient
+from .kling_client import KlingClient
+from .veo_client import VeoClient
+from .newapi_veo_client import NewAPIVeoClient
+from .grok_video_client import GrokVideoClient
+from .openai_client import OpenAIAPIClient

-__all__ = ['BaseAPIClient', 'GeminiAPIClient', 'GeminiFlashClient']
+__all__ = ['BaseAPIClient', 'GeminiAPIClient', 'GeminiFlashClient', 'SoraClient', 'KlingClient', 'VeoClient', 'NewAPIVeoClient', 'GrokVideoClient', 'OpenAIAPIClient']
@@ -0,0 +1,183 @@
+"""
+异步生图 Provider 抽象基类
+定义异步提交+轮询模式的统一接口，支持多种生图模型后端
+
+每个 Provider 封装一种 API 后端的通信协议：
+  - 如何提交任务（端点、请求体格式）
+  - 如何轮询状态（端点、状态字段语义）
+  - 如何解析结果（响应格式、图片提取方式）
+
+新增第三方生图模型时，只需实现此接口即可接入异步节点。
+"""
+
+from abc import ABC, abstractmethod
+from typing import Dict, List, Optional
+from PIL import Image
+
+
+class BaseAsyncImageProvider(ABC):
+    """异步生图 Provider 抽象基类"""
+
+    def __init__(self, api_key: str, proxy_url: Optional[str] = None):
+        self.api_key = api_key
+        self.proxy_url = proxy_url
+
+    # ========================================================================
+    # 必须实现的抽象方法
+    # ========================================================================
+
+    @property
+    @abstractmethod
+    def api_base_url(self) -> str:
+        """异步 API 的基础 URL，如 https://cf-api.o1key.com"""
+        ...
+
+    @abstractmethod
+    def get_submit_endpoint(self, model: str, resolution: str) -> str:
+        """获取提交任务的 API 端点路径（不含 base_url）"""
+        ...
+
+    @abstractmethod
+    def build_submit_body(
+        self,
+        prompt: str,
+        model: str,
+        resolution: str,
+        aspect_ratio: str,
+        images: Optional[List[Image.Image]] = None,
+        **kwargs
+    ) -> dict:
+        """构建提交任务的请求体"""
+        ...
+
+    @abstractmethod
+    def extract_task_id(self, response: dict) -> str:
+        """从提交响应中提取 task_id"""
+        ...
+
+    @abstractmethod
+    def extract_status(self, response: dict) -> str:
+        """从轮询响应中提取任务状态（如 SUBMITTED / IN_PROGRESS / SUCCESS / FAILURE）"""
+        ...
+
+    @abstractmethod
+    async def parse_result(
+        self,
+        result_data: dict,
+        session
+    ) -> List[Image.Image]:
+        """从任务完成后的 result data 中解析生成的图像列表"""
+        ...
+
+    @abstractmethod
+    def get_models(self) -> List[str]:
+        """获取此 Provider 支持的模型 ID 列表"""
+        ...
+
+    @abstractmethod
+    def get_model_aspect_ratios(self, model_id: str) -> List[str]:
+        """获取指定模型支持的宽高比"""
+        ...
+
+    @abstractmethod
+    def get_model_resolutions(self, model_id: str) -> List[str]:
+        """获取指定模型支持的分辨率"""
+        ...
+
+    # ========================================================================
+    # 可选的覆盖方法
+    # ========================================================================
+
+    def get_poll_endpoint(self, task_id: str) -> str:
+        """获取轮询任务状态的 API 端点路径（默认实现适用于 o1key 异步 API）"""
+        return f"/async/v1/tasks/{task_id}"
+
+    def get_headers(self) -> dict:
+        """获取 HTTP 请求头"""
+        return {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json"
+        }
+
+    def get_all_aspect_ratios(self) -> List[str]:
+        """获取所有模型支持的宽高比（去重合并）"""
+        seen = set()
+        result = []
+        for model_id in self.get_models():
+            for ratio in self.get_model_aspect_ratios(model_id):
+                if ratio not in seen:
+                    seen.add(ratio)
+                    result.append(ratio)
+        return result
+
+    def get_all_resolutions(self) -> List[str]:
+        """获取所有模型支持的分辨率（去重，按固定顺序排列）"""
+        _ORDER = ["512px", "1K", "2K", "4K"]
+        seen = set()
+        for model_id in self.get_models():
+            for res in self.get_model_resolutions(model_id):
+                seen.add(res)
+        return [r for r in _ORDER if r in seen]
+
+    def get_extra_inputs(self) -> dict:
+        """
+        返回此 Provider 特有的额外 ComfyUI 输入参数。
+        子类重写以声明 Provider 专有的选项（如 Google Search Grounding）。
+
+        Returns:
+            dict，格式与 ComfyUI INPUT_TYPES 的 optional 字段一致
+        """
+        return {}
+
+    def get_extra_kwargs(self, **kwargs) -> dict:
+        """
+        从 ComfyUI kwargs 中提取此 Provider 特有的参数，
+        转换为 build_submit_body 可接收的 kwargs。
+
+        子类重写以处理 Provider 专有参数。
+        """
+        return {}
+
+    def extract_progress(self, response: dict) -> Optional[float]:
+        """
+        从轮询响应中提取生成进度。
+
+        Args:
+            response: 轮询接口返回的完整响应字典
+
+        Returns:
+            0.0-1.0 之间的进度值，或 None 表示该响应不含进度信息
+        """
+        return None
+
+    def query_balance_sync(self) -> Optional[dict]:
+        """
+        同步查询账户余额（可选）。
+        返回 None 表示不支持。
+        """
+        return None
+
+    def format_balance_info(self, balance_data: dict) -> str:
+        """格式化余额信息为展示文本"""
+        return ""
+
+    # ========================================================================
+    # 工具方法
+    # ========================================================================
+
+    @staticmethod
+    def build_proxy_url(port: str) -> Optional[str]:
+        """
+        将端口号字符串转为 aiohttp 可用的 HTTP 代理 URL。
+        兼容 v2rayN (10808)、Clash Verge (7897) 等。
+
+        Args:
+            port: 用户填写的端口号，如 "7897"，空字符串返回 None
+
+        Returns:
+            代理 URL 或 None
+        """
+        port = (port or "").strip()
+        if not port or not port.isdigit():
+            return None
+        return f"http://127.0.0.1:{port}"
@@ -9,8 +9,14 @@ import threading
 from abc import ABC, abstractmethod
 from typing import Any, Callable, Dict, List, Optional

+import os
+import time
+
 import aiohttp

+from ..utils.http_error import HTTP_ERROR_MESSAGES, RETRYABLE_STATUS_CODES, _compute_delay, DEFAULT_MAX_RETRIES, DEFAULT_BASE_DELAY, DEFAULT_MAX_DELAY, DEFAULT_BACKOFF_FACTOR, get_friendly_message
+
+

 class BaseAPIClient(ABC):
    """
@@ -26,19 +32,21 @@ class BaseAPIClient(ABC):
        self,
        base_url: str,
        api_key: str,
-        max_request_size: int = 20 * 1024 * 1024
+        max_request_size: int = 100 * 1024 * 1024
    ):
        """
        初始化客户端
-        
+
        Args:
            base_url: API 基础 URL
            api_key: API 密钥
-            max_request_size: 最大请求体大小（字节），默认 20MB
+            max_request_size: 兼容参数；基类不再用它限制 JSON 请求体，
+                部分子类仍用它作为上传文件大小限制
        """
        self.base_url = base_url
        self.api_key = api_key
        self.max_request_size = max_request_size
+        self.proxy_url: Optional[str] = None  # 由节点在调用前注入，如 "http://127.0.0.1:7897"
    
    @abstractmethod
    def get_endpoint(self, **kwargs) -> str:
@@ -79,6 +87,15 @@ class BaseAPIClient(ABC):
        """
        pass
    
+    def _make_session(self) -> aiohttp.ClientSession:
+        """
+        创建统一的 aiohttp ClientSession，全局禁用 SSL 验证。
+        所有需要独立创建 session 的地方都应调用此方法，
+        避免因客户端系统缺少根证书导致 SSLCertVerificationError。
+        """
+        connector = aiohttp.TCPConnector(ssl=False, limit=0, limit_per_host=0)
+        return aiohttp.ClientSession(connector=connector, trust_env=False)
+
    def get_headers(self, use_bearer_token: bool = False) -> Dict[str, str]:
        """
        获取请求头
@@ -100,26 +117,19 @@ class BaseAPIClient(ABC):
                "Content-Type": "application/json"
            }
    
-    def check_request_size(self, request_body: Dict[str, Any]) -> None:
+    def get_http_error_message(self, status_code: int, error_message: str) -> Optional[str]:
        """
-        检查请求体大小是否超过限制
+        子类可重写：为指定 HTTP 状态码返回自定义错误文案。
+        若返回 None，则使用基类默认拼接文案。
        
        Args:
-            request_body: 请求体字典
+            status_code: HTTP 状态码（如 429、503）
+            error_message: API 返回的原始错误信息
        
-        Raises:
-            ValueError: 如果请求体超过限制
+        Returns:
+            自定义完整错误文案，或 None 表示使用默认
        """
-        request_json = json.dumps(request_body)
-        request_size = len(request_json.encode('utf-8'))
-        
-        if request_size > self.max_request_size:
-            size_mb = request_size / 1024 / 1024
-            limit_mb = self.max_request_size / 1024 / 1024
-            raise ValueError(
-                f"请求体大小 {size_mb:.2f}MB 超过限制 {limit_mb:.0f}MB，"
-                "请降低分辨率或减少图片数量"
-            )
+        return None
    
    async def request_async(
        self,
@@ -130,84 +140,155 @@ class BaseAPIClient(ABC):
        timeout: Optional[int] = None
    ) -> Dict[str, Any]:
        """
-        发送异步 HTTP 请求
-        
+        发送异步 HTTP 请求（带详细计时）
+
        Args:
            endpoint: API 端点
            request_body: 请求体
            session: aiohttp 会话（可选）
            use_bearer_token: 是否使用 Bearer Token 认证
-        
+            timeout: 超时时间（秒），默认 900 秒
+
        Returns:
            响应 JSON
-        
+
        Raises:
            RuntimeError: 请求失败时
+            InterruptProcessingException: 用户点击终止按钮时
        """
+        import time
+
+        # 尝试导入 ComfyUI 中断机制
+        try:
+            from comfy.model_management import processing_interrupted, InterruptProcessingException
+            _interrupt_available = True
+        except ImportError:
+            _interrupt_available = False
+
        url = f"{self.base_url}{endpoint}"
        headers = self.get_headers(use_bearer_token)
-        
-        # 检查请求大小
-        self.check_request_size(request_body)
-        
+
        close_session = False
        if session is None:
-            session = aiohttp.ClientSession()
+            session = self._make_session()
            close_session = True
-        
-        try:
-            # 设置超时
-            timeout_obj = aiohttp.ClientTimeout(total=timeout) if timeout else None
-            async with session.post(url, json=request_body, headers=headers, timeout=timeout_obj) as response:
+
+        # 设置请求超时：连接超时 30s，读取超时 900s（防止服务器出图后卡住）
+        _timeout_seconds = timeout if timeout is not None else 900
+        _aiohttp_timeout = aiohttp.ClientTimeout(
+            total=_timeout_seconds,
+            connect=30,
+            sock_read=_timeout_seconds
+        )
+
+        async def _do_request():
+            connect_start = time.time()
+            async with session.post(url, json=request_body, headers=headers, timeout=_aiohttp_timeout, proxy=self.proxy_url) as response:
+                connect_time = time.time() - connect_start
+
                if response.status != 200:
                    error_text = await response.text()
-                    
-                    # 针对常见错误状态码提供友好提示
-                    if response.status == 504:
-                        raise RuntimeError(
-                            f"API 请求超时 (504 Gateway Timeout)\n"
-                            f"原因：服务器响应超时或该端点暂时不可用\n"
-                            f"建议：\n"
-                            f"  - 尝试使用其他模型\n"
-                            f"  - 稍后重试\n"
-                            f"  - 降低分辨率或减少输入图像数量\n"
-                            f"详细错误: {error_text[:200]}"
-                        )
-                    elif response.status == 503:
-                        raise RuntimeError(
-                            f"服务暂时不可用 (503 Service Unavailable)\n"
-                            f"原因：模型服务过载或维护中\n"
-                            f"建议：\n"
-                            f"  - 稍后重试\n"
-                            f"  - 尝试使用其他模型"
-                        )
-                    elif response.status == 429:
-                        raise RuntimeError(
-                            f"请求频率超限 (429 Too Many Requests)\n"
-                            f"原因：API 配额用尽或请求过于频繁\n"
-                            f"建议：\n"
-                            f"  - 等待一段时间后重试\n"
-                            f"  - 检查 API 配额是否充足"
-                        )
-                    elif response.status == 404:
-                        raise RuntimeError(
-                            f"端点不存在 (404 Not Found)\n"
-                            f"原因：API 端点路径错误或模型不存在\n"
-                            f"建议：\n"
-                            f"  - 检查模型名称是否正确\n"
-                            f"  - 使用其他可用模型"
-                        )
-                    else:
-                        raise RuntimeError(
-                            f"API 请求失败 (状态码: {response.status}): {error_text}"
-                        )
-                
-                return await response.json()
-        
+                    # 返回状态码和错误文本，由外层处理重试
+                    return {"_error": True, "_status": response.status, "_text": error_text}
+
+                wait_start = time.time()
+                response_data = await response.json()
+                download_time = time.time() - wait_start
+
+                response_size = len(str(response_data))
+                if not isinstance(response_data, dict):
+                    response_data = {"data": response_data}
+
+                response_data["_timing"] = {
+                    "connect_time": connect_time,
+                    "download_time": download_time,
+                    "response_size": response_size
+                }
+                return response_data
+
+        async def _poll_interrupt():
+            """每 0.5s 轮询一次中断标志"""
+            while True:
+                await asyncio.sleep(0.5)
+                if processing_interrupted():
+                    return
+
+        try:
+            last_error_status = None
+            last_error_text = ""
+
+            for attempt in range(DEFAULT_MAX_RETRIES + 1):
+                if _interrupt_available:
+                    request_task = asyncio.ensure_future(_do_request())
+                    interrupt_task = asyncio.ensure_future(_poll_interrupt())
+
+                    done, pending = await asyncio.wait(
+                        [request_task, interrupt_task],
+                        return_when=asyncio.FIRST_COMPLETED
+                    )
+
+                    for t in pending:
+                        t.cancel()
+                        try:
+                            await t
+                        except (asyncio.CancelledError, Exception):
+                            pass
+
+                    if interrupt_task in done and request_task not in done:
+                        raise InterruptProcessingException()
+
+                    result = request_task.result()
+                else:
+                    result = await _do_request()
+
+                if isinstance(result, dict) and result.get("_error"):
+                    status = result["_status"]
+                    error_text = result["_text"]
+                    last_error_status = status
+                    last_error_text = error_text
+
+                    if status in RETRYABLE_STATUS_CODES and attempt < DEFAULT_MAX_RETRIES:
+                        friendly = HTTP_ERROR_MESSAGES.get(status, f"请求失败 ({status})")
+                        delay = _compute_delay(attempt, DEFAULT_BASE_DELAY, DEFAULT_MAX_DELAY, DEFAULT_BACKOFF_FACTOR)
+                        print(f"{friendly} {delay:.1f}s 后重试 ({attempt+1}/{DEFAULT_MAX_RETRIES})...")
+                        await asyncio.sleep(delay)
+                        continue
+
+                    if status in HTTP_ERROR_MESSAGES:
+                        raise RuntimeError(HTTP_ERROR_MESSAGES[status])
+                    raise RuntimeError(get_friendly_message(status, error_text))
+
+                return result
+
+            if last_error_status and last_error_status in HTTP_ERROR_MESSAGES:
+                raise RuntimeError(HTTP_ERROR_MESSAGES[last_error_status])
+            raise RuntimeError(get_friendly_message(last_error_status or 0, last_error_text))
+
+        except InterruptProcessingException:
+            raise
+
+        except aiohttp.ServerTimeoutError as e:
+            raise RuntimeError(
+                f"请求超时！等待服务器响应超过 {_timeout_seconds} 秒。\n"
+                f"服务器可能仍在生成图片，请稍后重试，或检查网络连接。"
+            ) from e
+
+        except aiohttp.ClientConnectorError as e:
+            raise RuntimeError(
+                f"无法连接到服务器：{str(e)}\n"
+                f"请检查网络连接是否正常。"
+            ) from e
+
+        except asyncio.TimeoutError as e:
+            raise RuntimeError(
+                f"请求超时！等待服务器响应超过 {_timeout_seconds} 秒。\n"
+                f"服务器可能仍在生成图片，请稍后重试，或检查网络连接。"
+            ) from e
+
        finally:
            if close_session:
                await session.close()
-    
+
    async def request_get_async(
        self,
        endpoint: str,
@@ -222,6 +303,7 @@ class BaseAPIClient(ABC):
            endpoint: API 端点
            session: aiohttp 会话（可选）
            use_bearer_token: 是否使用 Bearer Token 认证（默认为 True）
+            timeout: 超时时间（秒）- 已废弃，由服务器端控制
        
        Returns:
            响应 JSON
@@ -234,41 +316,67 @@ class BaseAPIClient(ABC):
        
        close_session = False
        if session is None:
-            session = aiohttp.ClientSession()
+            session = self._make_session()
            close_session = True
-        
+
        try:
-            # 设置超时
-            timeout_obj = aiohttp.ClientTimeout(total=timeout) if timeout else None
-            async with session.get(url, headers=headers, timeout=timeout_obj) as response:
+            _get_start = time.time()
+            async with session.get(url, headers=headers) as response:
+                _get_elapsed = time.time() - _get_start
                if response.status != 200:
                    error_text = await response.text()
                    
+                    # 尝试解析 JSON 错误信息，提取关键内容
+                    error_message = error_text
+                    try:
+                        error_json = json.loads(error_text)
+                        # 尝试从多个常见位置提取错误信息
+                        if "error" in error_json:
+                            if isinstance(error_json["error"], dict):
+                                error_message = error_json["error"].get("message", error_text)
+                            else:
+                                error_message = str(error_json["error"])
+                        elif "message" in error_json:
+                            error_message = error_json["message"]
+                    except:
+                        # 如果不是 JSON，使用原始文本
+                        pass
+                    
                    # 针对常见错误状态码提供友好提示
-                    if response.status == 504:
+                    if response.status == 400:
                        raise RuntimeError(
-                            f"API 请求超时 (504 Gateway Timeout)\n"
-                            f"原因：服务器响应超时或该端点暂时不可用\n"
-                            f"建议：稍后重试"
+                            f"请求参数错误 (400 Bad Request)\n"
+                            f"API 返回错误：{error_message}\n"
+                            f"建议：检查请求参数"
                        )
-                    elif response.status == 503:
+                    elif response.status == 401:
                        raise RuntimeError(
-                            f"服务暂时不可用 (503 Service Unavailable)\n"
-                            f"原因：服务过载或维护中\n"
-                            f"建议：稍后重试"
+                            f"认证失败 (401 Unauthorized)\n"
+                            f"API 返回错误：{error_message}\n"
+                            f"建议：检查 API 密钥"
                        )
                    elif response.status == 429:
-                        raise RuntimeError(
-                            f"请求频率超限 (429 Too Many Requests)\n"
-                            f"原因：API 配额用尽或请求过于频繁\n"
-                            f"建议：等待一段时间后重试"
-                        )
+                        custom = self.get_http_error_message(429, error_message)
+                        if custom is not None:
+                            raise RuntimeError(custom)
+                        raise RuntimeError(HTTP_ERROR_MESSAGES[429])
+                    elif response.status == 503:
+                        custom = self.get_http_error_message(503, error_message)
+                        if custom is not None:
+                            raise RuntimeError(custom)
+                        raise RuntimeError(HTTP_ERROR_MESSAGES[503])
+                    elif response.status == 504:
+                        raise RuntimeError(HTTP_ERROR_MESSAGES[504])
+                    elif response.status == 502:
+                        raise RuntimeError(HTTP_ERROR_MESSAGES[502])
                    else:
                        raise RuntimeError(
-                            f"API 请求失败 (状态码: {response.status}): {error_text}"
+                            f"API 请求失败 (状态码: {response.status})\n"
+                            f"API 返回错误：{error_message}"
                        )
                
-                return await response.json()
+                _resp_data = await response.json()
+                return _resp_data
        
        finally:
            if close_session:
@@ -294,9 +402,7 @@ class BaseAPIClient(ABC):
        total = len(requests)
        
        # 创建无限制的连接器
-        connector = aiohttp.TCPConnector(limit=0, limit_per_host=0)
-        
-        async with aiohttp.ClientSession(connector=connector) as session:
+        async with self._make_session() as session:
            tasks = []
            
            for req in requests:
@@ -366,3 +472,58 @@ class BaseAPIClient(ABC):
            raise RuntimeError("异步任务未返回结果")
        
        return result_container[0]
+    
+    async def query_balance_async(self) -> Dict[str, Any]:
+        """
+        异步查询账户余额
+        
+        Returns:
+            余额信息字典，包含 name、total_available 等字段
+        
+        Raises:
+            RuntimeError: 查询失败时
+        """
+        endpoint = "/api/usage/token"
+        response = await self.request_get_async(endpoint, use_bearer_token=True)
+        
+        if not response.get("code"):
+            raise RuntimeError("余额查询响应格式错误")
+        
+        data = response.get("data", {})
+        return data
+    
+    def query_balance_sync(self) -> Dict[str, Any]:
+        """
+        同步查询账户余额（用于 ComfyUI 节点）
+        
+        Returns:
+            余额信息字典
+        
+        Raises:
+            RuntimeError: 查询失败时
+        """
+        coro = self.query_balance_async()
+        return self.run_async_in_thread(coro)
+    
+    def format_balance_info(self, balance_data: Dict[str, Any]) -> str:
+        """
+        格式化余额信息为展示文本
+        
+        Args:
+            balance_data: 余额信息字典
+        
+        Returns:
+            格式化文本，如 "当前余额：100.00 | API：xxx"
+        
+        Example:
+            >>> data = {"name": "test-api", "total_available": 50000000}
+            >>> client.format_balance_info(data)
+            '当前余额：100.00 | API：test-api'
+        """
+        api_name = balance_data.get("name", "未知")
+        total_available = balance_data.get("total_available", 0)
+        
+        # 实际显示余额 = total_available / 500000，单位：美元
+        balance_in_dollars = total_available / 500000
+        
+        return f"当前余额：{balance_in_dollars:.2f} | API：{api_name}"
@@ -0,0 +1,358 @@
+"""
+豆包生图 API 客户端
+端点：POST /v1/images/generations/
+兼容 new-api 透传格式（OpenAI images/generations 兼容）
+
+设计原则：
+  - 发送完整正确的请求体，new-api 丢弃字段是其侧问题
+  - 响应永远是同步 JSON（new-api 强制 stream=false）
+  - 图像输入以 data:image/png;base64,... 格式内联传递
+"""
+
+import asyncio
+import json
+import time
+from io import BytesIO
+from concurrent.futures import ThreadPoolExecutor
+from typing import List, Optional, Union
+
+import aiohttp
+from PIL import Image
+
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.http_error import RETRYABLE_STATUS_CODES, HTTP_ERROR_MESSAGES, _compute_delay, DEFAULT_MAX_RETRIES, DEFAULT_BASE_DELAY, DEFAULT_MAX_DELAY, DEFAULT_BACKOFF_FACTOR
+
+
+# ── 固定端点 ──────────────────────────────────────────────────────────────────
+_ENDPOINT = "/v1/images/generations/"
+
+# ── 轮询 / 请求超时 ───────────────────────────────────────────────────────────
+_REQUEST_TIMEOUT = 300   # 单次请求超时秒数（豆包图像生成最长约 60s）
+
+
+class DoubaoImageClient:
+    """
+    豆包生图客户端（new-api 原生 OpenAI 兼容格式）
+
+    new-api 兼容性说明（基于源码分析）：
+      ✅ 透传：model / prompt / size / response_format / watermark / image
+      ❌ 丢弃：seed / sequential_image_generation / sequential_image_generation_options
+              （进入 Extra map，但 MarshalJSON 中合并代码被注释）
+      ❌ 强制：stream 硬编码 false，图像接口无流式处理
+      ❌ 未实现：/v1/files 文件上传（501）
+
+    节点仍发送完整字段，待 new-api 修复后自动生效。
+    """
+
+    def __init__(self):
+        self.api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        self.base_url = get_api_base_url()
+
+    # ── 认证头 ────────────────────────────────────────────────────────────────
+
+    def _headers(self) -> dict:
+        return {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+
+    # ── 图像字段构建 ──────────────────────────────────────────────────────────
+
+    def _tensor_to_image_field(self, tensor) -> Union[str, List[str]]:
+        """
+        ComfyUI IMAGE tensor → API image 字段值
+
+        单张返回字符串，多张返回字符串列表，格式：
+            data:image/png;base64,<base64数据>
+        """
+        pil_images = tensor_to_pil(tensor)
+        data_urls = []
+        for img in pil_images:
+            b64 = encode_image_to_base64(img, format="PNG")
+            data_urls.append(f"data:image/png;base64,{b64}")
+
+        return data_urls[0] if len(data_urls) == 1 else data_urls
+
+    # ── 请求体构建 ────────────────────────────────────────────────────────────
+
+    def _build_body(
+        self,
+        model: str,
+        prompt: str,
+        size: str,
+        seed: int,
+        sequential_image_generation: str,
+        max_images: int,
+        image_field=None,          # str | list[str] | None
+    ) -> dict:
+        """
+        构建完整请求体。
+
+        字段说明（对照官方示例）：
+          - response_format: 固定 "url"（new-api 原样透传给豆包）
+          - watermark: 固定 False（UI 已移除该参数）
+          - stream: 固定 False（new-api 强制非流式，此字段不被读取，仅显式注明）
+          - sequential_image_generation_options: 仅 sequential=auto 时发送
+        """
+        body = {
+            "model": model,
+            "prompt": prompt,
+            "size": size,
+            "response_format": "url",
+            "watermark": False,
+            "seed": seed,
+            "sequential_image_generation": sequential_image_generation,
+        }
+
+        # 仅 auto 模式才发送 max_images 选项
+        if sequential_image_generation == "auto":
+            body["sequential_image_generation_options"] = {
+                "max_images": max_images
+            }
+
+        # 图像输入（图生图）
+        if image_field is not None:
+            body["image"] = image_field
+
+        return body
+
+    # ── 图像下载 ──────────────────────────────────────────────────────────────
+
+    async def _download_image(
+        self,
+        url: str,
+        session: aiohttp.ClientSession,
+    ) -> Image.Image:
+        """从 URL 下载图像，返回 PIL.Image。"""
+        async with session.get(url, allow_redirects=True) as resp:
+            if resp.status != 200:
+                raise RuntimeError(
+                    f"图像下载失败，HTTP {resp.status}，URL: {url}"
+                )
+            data = await resp.read()
+
+        try:
+            img = Image.open(BytesIO(data)).convert("RGB")
+        except Exception as e:
+            raise RuntimeError(f"图像解码失败: {e}")
+        return img
+
+    # ── 响应解析 ──────────────────────────────────────────────────────────────
+
+    async def _parse_response(
+        self,
+        resp_json: dict,
+        session: aiohttp.ClientSession,
+    ) -> List[Image.Image]:
+        """
+        解析 /v1/images/generations 响应，返回 PIL.Image 列表。
+
+        期望格式（new-api 原样透传豆包响应）：
+        {
+            "created": 1234567890,
+            "data": [
+                {"url": "https://..."},
+                {"url": "https://..."}
+            ]
+        }
+
+        兼容 b64_json 字段（豆包理论上也支持）。
+        """
+        # 检查 API 层级错误
+        if "error" in resp_json:
+            err = resp_json["error"]
+            if isinstance(err, dict):
+                msg = err.get("message") or err.get("msg") or json.dumps(err, ensure_ascii=False)
+            else:
+                msg = str(err)
+            raise RuntimeError(f"API 返回错误: {msg}")
+
+        data_list = resp_json.get("data")
+        if not data_list:
+            raise RuntimeError(
+                f"API 响应中未找到 data 字段，完整响应：\n"
+                f"{json.dumps(resp_json, ensure_ascii=False, indent=2)}"
+            )
+
+        images: List[Image.Image] = []
+
+        for idx, item in enumerate(data_list):
+            url = item.get("url", "")
+            b64 = item.get("b64_json", "")
+
+            if url and url.startswith("http"):
+                # 优先使用 URL 模式
+                img = await self._download_image(url, session)
+                images.append(img)
+                print(f"[豆包生图] 第 {idx + 1} 张下载完成 ({img.size[0]}×{img.size[1]})")
+
+            elif b64:
+                # 回退到 base64 模式
+                import base64 as _b64
+                try:
+                    img_data = _b64.b64decode(b64)
+                    img = Image.open(BytesIO(img_data)).convert("RGB")
+                    images.append(img)
+                    print(f"[豆包生图] 第 {idx + 1} 张 base64 解码完成 ({img.size[0]}×{img.size[1]})")
+                except Exception as e:
+                    raise RuntimeError(f"第 {idx + 1} 张 base64 解码失败: {e}")
+
+            else:
+                print(f"[豆包生图] 警告：第 {idx + 1} 条数据既无 url 也无 b64_json，已跳过")
+
+        return images
+
+    # ── 核心异步生成方法 ──────────────────────────────────────────────────────
+
+    async def _generate_async(
+        self,
+        model: str,
+        prompt: str,
+        size: str,
+        seed: int,
+        sequential_image_generation: str,
+        max_images: int,
+        image_tensor=None,
+    ) -> List[Image.Image]:
+        """
+        异步完整流程：构建请求 → POST → 解析 → 下载图像。
+        """
+        # 1. 构建 image 字段
+        image_field = None
+        if image_tensor is not None:
+            image_field = self._tensor_to_image_field(image_tensor)
+            n_imgs = len(image_field) if isinstance(image_field, list) else 1
+            print(f"[豆包生图] 图生图模式，参考图 {n_imgs} 张")
+        else:
+            print(f"[豆包生图] 文生图模式")
+
+        # 2. 构建请求体
+        body = self._build_body(
+            model=model,
+            prompt=prompt,
+            size=size,
+            seed=seed,
+            sequential_image_generation=sequential_image_generation,
+            max_images=max_images,
+            image_field=image_field,
+        )
+
+        url = f"{self.base_url}{_ENDPOINT}"
+        print(f"[豆包生图] 提交请求 → {model} | {size}")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        timeout = aiohttp.ClientTimeout(total=_REQUEST_TIMEOUT)
+
+        async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
+
+            # 3. 发送 POST 请求（带退避重试）
+            last_status = None
+            for attempt in range(DEFAULT_MAX_RETRIES + 1):
+                t0 = time.time()
+                async with session.post(
+                    url,
+                    json=body,
+                    headers=self._headers(),
+                ) as resp:
+                    elapsed_req = time.time() - t0
+                    text = await resp.text()
+
+                    if resp.status != 200:
+                        last_status = resp.status
+                        if resp.status in RETRYABLE_STATUS_CODES and attempt < DEFAULT_MAX_RETRIES:
+                            friendly = HTTP_ERROR_MESSAGES.get(resp.status)
+                            delay = _compute_delay(attempt, DEFAULT_BASE_DELAY, DEFAULT_MAX_DELAY, DEFAULT_BACKOFF_FACTOR)
+                            print(f"[豆包生图] {friendly} {delay:.1f}s 后重试 ({attempt+1}/{DEFAULT_MAX_RETRIES})...")
+                            await asyncio.sleep(delay)
+                            continue
+                        if resp.status in HTTP_ERROR_MESSAGES:
+                            raise RuntimeError(HTTP_ERROR_MESSAGES[resp.status])
+                        try:
+                            err_json = json.loads(text)
+                            err_obj = err_json.get("error", {})
+                            if isinstance(err_obj, dict):
+                                msg = (
+                                    err_obj.get("message")
+                                    or err_obj.get("msg")
+                                    or text
+                                )
+                            else:
+                                msg = str(err_obj) or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(
+                            f"请求失败 HTTP {resp.status}: {msg}"
+                        )
+
+                    try:
+                        resp_json = json.loads(text)
+                    except Exception:
+                        raise RuntimeError(f"响应 JSON 解析失败，原始内容：{text[:500]}")
+
+                break
+            else:
+                if last_status and last_status in HTTP_ERROR_MESSAGES:
+                    raise RuntimeError(HTTP_ERROR_MESSAGES[last_status])
+                raise RuntimeError(f"请求失败: 重试 {DEFAULT_MAX_RETRIES} 次后仍然失败")
+
+            print(f"[豆包生图] API 响应耗时 {elapsed_req:.1f}s，开始下载图像...")
+
+            # 4. 解析响应 & 下载图像（session 复用）
+            images = await self._parse_response(resp_json, session)
+
+        return images
+
+    # ── 同步入口（供 ComfyUI 节点调用）──────────────────────────────────────
+
+    def generate_sync(
+        self,
+        model: str,
+        prompt: str,
+        size: str,
+        seed: int,
+        sequential_image_generation: str,
+        max_images: int,
+        image_tensor=None,
+    ) -> List[Image.Image]:
+        """
+        同步生成接口（在独立线程中运行事件循环，避免与 ComfyUI 主循环冲突）。
+
+        Args:
+            model:                       模型 ID
+            prompt:                      提示词
+            size:                        尺寸字符串，如 "2048x2048"
+            seed:                        随机种子
+            sequential_image_generation: "disabled" | "auto"
+            max_images:                  最大图片数（auto 模式生效）
+            image_tensor:                ComfyUI IMAGE tensor（可选，图生图用）
+
+        Returns:
+            List[PIL.Image]
+        """
+        coro = self._generate_async(
+            model=model,
+            prompt=prompt,
+            size=size,
+            seed=seed,
+            sequential_image_generation=sequential_image_generation,
+            max_images=max_images,
+            image_tensor=image_tensor,
+        )
+
+        def _run():
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            try:
+                return loop.run_until_complete(coro)
+            finally:
+                loop.close()
+
+        with ThreadPoolExecutor(max_workers=1) as executor:
+            future = executor.submit(_run)
+            try:
+                return future.result(timeout=_REQUEST_TIMEOUT + 30)
+            except TimeoutError:
+                raise RuntimeError(
+                    f"豆包生图超时（>{_REQUEST_TIMEOUT}s），请检查网络或稍后重试"
+                )
@@ -0,0 +1,199 @@
+"""
+Flux 图像编辑 API 客户端
+通过 vip.o1key.com 调用 Flux2 图像编辑 + SeedVR2 超分辨率服务
+
+工作流程：
+1. submit_task  → POST /v1/images/edits (multipart/form-data 提交主图+参考图+提示词)
+2. poll_result  → GET  /v1/images/edits/{task_id} (直连容器轮询)
+"""
+
+import base64
+import time
+from typing import Optional
+
+import requests
+
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..utils.http_error import HTTP_ERROR_MESSAGES
+
+
+# 显示名 → 实际请求值的映射
+SIZE_DISPLAY_MAP = {
+    "2K": "2048",
+    "4K": "4096",
+}
+
+POLL_BASE_URL = "https://xrrh7tn08tfgwa8w-8188.container.x-gpu.com"
+
+
+class FluxEditClient:
+    """
+    Flux 图像编辑客户端
+
+    对接 vip.o1key.com 上的 /v1/images/edits 接口，
+    将图像编辑+超分辨率任务提交到远程服务器执行。
+    """
+
+    SUBMIT_ENDPOINT = "/v1/images/edits"
+    STATUS_ENDPOINT = "/v1/images/edits/{task_id}"
+
+    DEFAULT_POLL_INTERVAL = 15  # 秒
+
+    def __init__(self):
+        self.api_key = get_api_key_or_raise()
+        self.base_url = get_api_base_url()
+
+    # ------------------------------------------------------------------
+    # 同步方法（供 ComfyUI 节点调用）
+    # ------------------------------------------------------------------
+
+    def submit_and_wait(
+        self,
+        image_bytes: bytes,
+        mask_bytes: bytes,
+        prompt: str,
+        size: str = "4K",
+        poll_interval: int = DEFAULT_POLL_INTERVAL,
+        progress_callback=None,
+    ) -> bytes:
+        """
+        提交任务并同步等待结果（阻塞直到完成）
+
+        Args:
+            image_bytes: 主图二进制数据
+            mask_bytes: 参考图二进制数据
+            prompt: 编辑提示词
+            size: 分辨率显示名 ("2K" 或 "4K")
+            poll_interval: 轮询间隔（秒）
+            progress_callback: 进度回调 fn(status_str)
+
+        Returns:
+            结果图像的二进制数据
+
+        Raises:
+            RuntimeError: 任务失败
+        """
+        size_value = SIZE_DISPLAY_MAP.get(size, size)
+
+        task_id = self._submit_task_sync(image_bytes, mask_bytes, prompt, size_value)
+        if progress_callback:
+            progress_callback(f"任务已提交: {task_id[:8]}...")
+
+        # 2. 轮询等待（直连容器）
+        return self._poll_result_sync(
+            task_id, poll_interval, progress_callback
+        )
+
+    def _submit_task_sync(
+        self,
+        image_bytes: bytes,
+        mask_bytes: bytes,
+        prompt: str,
+        size: str,
+    ) -> str:
+        """同步提交任务，返回 task_id"""
+        url = f"{self.base_url}{self.SUBMIT_ENDPOINT}"
+        headers = {"Authorization": f"Bearer {self.api_key}"}
+
+        files = {
+            "image": ("image.jpg", image_bytes, "image/jpeg"),
+            "mask": ("mask.jpg", mask_bytes, "image/jpeg"),
+        }
+        data = {
+            "prompt": prompt,
+            "size": size,
+            "model": "flux2-fp8-dualr",
+        }
+
+        try:
+            resp = requests.post(url, files=files, data=data, headers=headers, timeout=60)
+        except requests.exceptions.Timeout:
+            raise RuntimeError("提交任务超时，请检查网络连接")
+        except requests.exceptions.ConnectionError:
+            raise RuntimeError("无法连接到服务器，请检查网络或服务器地址")
+
+        if resp.status_code != 200:
+            if resp.status_code in HTTP_ERROR_MESSAGES:
+                raise RuntimeError(HTTP_ERROR_MESSAGES[resp.status_code])
+            raise RuntimeError(
+                f"提交任务失败 (HTTP {resp.status_code})\n"
+                f"响应: {resp.text[:500]}"
+            )
+
+        result = resp.json()
+        task_id = result.get("id")
+        if not task_id:
+            raise RuntimeError(f"服务器返回异常: 未获取到任务ID\n{result}")
+
+        return task_id
+
+    def _poll_result_sync(
+        self,
+        task_id: str,
+        poll_interval: int,
+        progress_callback=None,
+    ) -> bytes:
+        """同步轮询任务状态（直连容器），返回结果图像二进制"""
+        url = f"{POLL_BASE_URL}{self.STATUS_ENDPOINT.format(task_id=task_id)}"
+
+        start_time = time.time()
+        last_status = None
+
+        while True:
+            elapsed = time.time() - start_time
+
+            try:
+                resp = requests.get(url, timeout=30)
+            except requests.exceptions.ConnectionError:
+                raise RuntimeError("轮询时无法连接到服务器，请检查网络")
+
+            if resp.status_code != 200:
+                raise RuntimeError(
+                    f"查询任务状态失败 (HTTP {resp.status_code})\n"
+                    f"响应: {resp.text[:500]}"
+                )
+
+            result = resp.json()
+            status = result.get("status", "unknown")
+
+            # 状态变化时打印日志
+            if status != last_status:
+                elapsed_str = f"{elapsed:.0f}s"
+                print(f"Flux Edit: [{elapsed_str}] 任务 {task_id[:8]}... → {status}")
+                last_status = status
+
+            if progress_callback:
+                elapsed_str = f"{elapsed:.0f}s"
+                status_desc = {
+                    "pending": "排队中",
+                    "processing": "处理中",
+                    "generating": "生图中，请耐心等待，预计耗时140s左右",
+                }.get(status, status)
+                progress_callback(f"{status_desc} (当前进度：{elapsed_str})")
+
+            if status == "completed":
+                # 解码 base64 图像
+                b64_data = result.get("result")
+                if not b64_data:
+                    raise RuntimeError("任务完成但未返回图像数据")
+                return base64.b64decode(b64_data)
+
+            elif status == "failed":
+                error_msg = result.get("error", "未知错误")
+                raise RuntimeError(
+                    f"图像编辑任务失败\n"
+                    f"错误: {error_msg}"
+                )
+
+            elif status in ("not_found",):
+                raise RuntimeError(
+                    f"任务未找到: {task_id}\n"
+                    f"可能已被清理或 ID 无效"
+                )
+
+            # 继续等待
+            time.sleep(poll_interval)
+
+    def query_balance_sync(self) -> dict:
+        """查询余额（兼容现有节点的 finally 块调用）"""
+        return {"name": "flux-edit", "total_available": 0}
@@ -0,0 +1,194 @@
+"""
+Gemini 异步生图 Provider
+通过 cf-api.o1key.com 的异步提交+轮询接口调用 Gemini 图像生成模型
+
+协议说明：
+  - 提交：POST {base}/async{gemini_endpoint}?image_format=url
+  - 轮询：GET  {base}/async/v1/tasks/{task_id}
+  - 结果：可能直接返回 image_url，也可能返回 Gemini 标准 candidates 格式
+"""
+
+from io import BytesIO
+from typing import Dict, List, Optional
+from PIL import Image
+
+from .base_async_provider import BaseAsyncImageProvider
+from .gemini_client import GeminiAPIClient
+from ..utils.config import get_async_api_base_url, get_api_key_or_raise
+from ..models_config import (
+    get_enabled_models,
+    get_model_supported_aspect_ratios,
+    get_model_supported_resolutions,
+)
+
+
+class GeminiAsyncImageProvider(BaseAsyncImageProvider):
+    """
+    Gemini 异步生图 Provider
+
+    委托 GeminiAPIClient 处理：
+      - 端点构造（get_endpoint）
+      - 请求体构建（build_request_body）
+      - 响应解析（parse_response_async）
+    """
+
+    def __init__(self, api_key: str = None, proxy_url: str = None):
+        if api_key is None:
+            api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        super().__init__(api_key=api_key, proxy_url=proxy_url)
+        self._client = GeminiAPIClient(api_key=api_key)
+
+    # ========================================================================
+    # 抽象方法实现
+    # ========================================================================
+
+    @property
+    def api_base_url(self) -> str:
+        return getattr(self, '_route_base_url', None) or get_async_api_base_url()
+
+    def get_submit_endpoint(self, model: str, resolution: str) -> str:
+        gemini_endpoint = self._client.get_endpoint(
+            model=model, resolution=resolution, image_format="url"
+        )
+        base = gemini_endpoint.split("?")[0]
+        async_endpoint = f"/async{base}"
+        if "?" in gemini_endpoint:
+            async_endpoint += "?" + gemini_endpoint.split("?", 1)[1]
+        return async_endpoint
+
+    def build_submit_body(
+        self,
+        prompt: str,
+        model: str,
+        resolution: str,
+        aspect_ratio: str,
+        images: Optional[List[Image.Image]] = None,
+        **kwargs
+    ) -> dict:
+        return self._client.build_request_body(
+            prompt=prompt,
+            images=images,
+            aspect_ratio=aspect_ratio,
+            resolution=resolution,
+            enable_grounding=kwargs.get("enable_grounding", False),
+            enable_image_search=kwargs.get("enable_image_search", False),
+            image_compression=getattr(self, "image_compression", None),
+            thinking_level=kwargs.get("thinking_level"),
+            request_log_enabled=False,
+        )
+
+    def extract_task_id(self, response: dict) -> str:
+        task_id = response.get("task_id")
+        if not task_id:
+            raise RuntimeError(f"提交响应中未找到 task_id: {response}")
+        return task_id
+
+    def extract_status(self, response: dict) -> str:
+        return response.get("status", "UNKNOWN")
+
+    async def parse_result(self, result_data: dict, session) -> List[Image.Image]:
+        images = result_data.get("images") if isinstance(result_data, dict) else None
+        if isinstance(images, list) and images:
+            parsed = []
+            for item in images:
+                if not isinstance(item, dict):
+                    continue
+                image_url = item.get("url") or item.get("image_url")
+                if image_url:
+                    async with session.get(image_url) as img_resp:
+                        if img_resp.status == 200:
+                            img_bytes = await img_resp.read()
+                            parsed.append(Image.open(BytesIO(img_bytes)).convert("RGB"))
+                        else:
+                            raise RuntimeError(f"下载图片失败 ({img_resp.status}): {image_url}")
+            if parsed:
+                return parsed
+
+        # 异步接口可能直接返回 image_url
+        image_url = result_data.get("image_url", "") if isinstance(result_data, dict) else ""
+        if image_url:
+            async with session.get(image_url) as img_resp:
+                if img_resp.status == 200:
+                    img_bytes = await img_resp.read()
+                    return [Image.open(BytesIO(img_bytes))]
+                raise RuntimeError(f"下载图片失败 ({img_resp.status}): {image_url}")
+
+        # 否则按 Gemini 标准格式解析
+        images_list, _ = await self._client.parse_response_async(result_data, session=session)
+        return images_list
+
+    def get_models(self) -> List[str]:
+        return get_enabled_models()
+
+    def get_model_aspect_ratios(self, model_id: str) -> List[str]:
+        return get_model_supported_aspect_ratios(model_id)
+
+    def get_model_resolutions(self, model_id: str) -> List[str]:
+        return get_model_supported_resolutions(model_id)
+
+    # ========================================================================
+    # 可选方法覆盖
+    # ========================================================================
+
+    def get_extra_inputs(self) -> dict:
+        """Gemini 专有：Google Search Grounding"""
+        return {
+            "联网功能": (["关闭", "打开"], {"default": "关闭"}),
+        }
+
+    def get_extra_kwargs(self, **kwargs) -> dict:
+        return {
+            "enable_grounding": kwargs.pop("联网功能", "关闭") == "打开",
+        }
+
+    def extract_progress(self, response: dict) -> Optional[float]:
+        """从轮询响应中提取进度（0.0-1.0）"""
+
+        def _coerce(val) -> Optional[float]:
+            if val is None or isinstance(val, bool):
+                return None
+            if isinstance(val, (int, float)):
+                progress = float(val)
+            elif isinstance(val, str):
+                text = val.strip()
+                if not text:
+                    return None
+                has_percent_suffix = text.endswith("%")
+                if has_percent_suffix:
+                    text = text[:-1].strip()
+                try:
+                    progress = float(text)
+                except ValueError:
+                    return None
+                if has_percent_suffix:
+                    progress /= 100.0
+            else:
+                return None
+            if progress > 1.0:
+                progress /= 100.0
+            return max(0.0, min(progress, 1.0))
+
+        # 直接字段：progress / percentage
+        for field in ("progress", "percentage", "percent"):
+            progress = _coerce(response.get(field))
+            if progress is not None:
+                return progress
+
+        # 嵌套字段：progressInfo / progress_info
+        progress_info = response.get("progressInfo") or response.get("progress_info")
+        if isinstance(progress_info, dict):
+            for field in ("progress", "percentage", "percent"):
+                progress = _coerce(progress_info.get(field))
+                if progress is not None:
+                    return progress
+
+        return None
+
+    def query_balance_sync(self) -> Optional[dict]:
+        try:
+            return self._client.query_balance_sync()
+        except Exception:
+            return None
+
+    def format_balance_info(self, balance_data: dict) -> str:
+        return self._client.format_balance_info(balance_data)
@@ -8,15 +8,15 @@ from typing import Any, Dict, List, Optional

 import aiohttp

-from ..utils.config import get_api_key_or_raise
-from ..models_config import get_flash_model_endpoint, get_enabled_flash_models
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..models_config import (
+    get_flash_model_endpoint,
+    get_enabled_flash_models,
+    get_flash_model_thinking_level_value,
+)
 from .base_client import BaseAPIClient


-# API 基础配置
-API_BASE_URL = "https://api.o1key.com"
-
-
 class GeminiFlashClient(BaseAPIClient):
    """
    Gemini Flash API 客户端
@@ -24,8 +24,7 @@ class GeminiFlashClient(BaseAPIClient):
    
    特点：
    - 支持图片和视频输入
-    - 支持系统指令
-    - 支持不同思考深度
+    - 支持动态思考等级端点（不思考/低/中/高）
    """
    
    def __init__(self, api_key: Optional[str] = None):
@@ -39,46 +38,65 @@ class GeminiFlashClient(BaseAPIClient):
            api_key = get_api_key_or_raise("O1KEY_API_KEY")
        
        super().__init__(
-            base_url=API_BASE_URL,
+            base_url=get_api_base_url(),
            api_key=api_key,
-            max_request_size=20 * 1024 * 1024  # 20MB
+            max_request_size=100 * 1024 * 1024  # 100MB
        )
    
    def get_endpoint(
-        self, 
+        self,
        model: str = "gemini-3-flash-preview",
-        thinking_depth: str = "不思考", 
        **kwargs
    ) -> str:
        """
-        根据模型和思考深度获取 API 端点
+        获取模型的 API 端点
        
        Args:
            model: 模型名称
-            thinking_depth: 思考深度 ("不思考" 或 "高")
        
        Returns:
            API 端点路径
        """
-        endpoint = get_flash_model_endpoint(model, thinking_depth)
+        endpoint = get_flash_model_endpoint(model)
        
        if endpoint is None:
-            # 回退到默认端点
+            # 回退到第一个启用的模型端点
            default_models = get_enabled_flash_models()
            if default_models:
-                endpoint = get_flash_model_endpoint(default_models[0], thinking_depth)
+                endpoint = get_flash_model_endpoint(default_models[0])
        
        if endpoint is None:
-            raise ValueError(f"无法获取模型 '{model}' 的端点 (思考深度: {thinking_depth})")
+            raise ValueError(f"无法获取模型 '{model}' 的端点")
        
        return endpoint
    
+    def get_http_error_message(self, status_code: int, error_message: str) -> Optional[str]:
+        """Gemini 请求 429/503 时返回图中约定的多行错误框文案。"""
+        if status_code == 429:
+            return (
+                "莫慌!该模型暂时超出速率限制啦\n"
+                "解决方案如下(任意一种):\n"
+                "1.切换当前模型\n"
+                "2.前往后台,修改令牌分组"
+            )
+        if status_code == 503:
+            return (
+                "警报!谷歌服务器当前过载!\n"
+                "解决方案如下:\n"
+                "1.摸会儿鱼吧,我也没办法,谷歌会尽快恢复,嘿嘿~\n"
+                "2.切换其他模型\n"
+                "3.前往后台,修改令牌分组"
+            )
+        return None
+    
    def build_request_body(
        self,
        prompt: str = "",
-        system_instruction: Optional[str] = None,
+        model: str = "gemini-3-flash-preview",
+        thinking_level: str = "不思考",
        image_data: Optional[List[Dict[str, str]]] = None,
        video_data: Optional[Dict[str, str]] = None,
+        document_data: Optional[Dict[str, str]] = None,
        **kwargs
    ) -> Dict[str, Any]:
        """
@@ -86,9 +104,11 @@ class GeminiFlashClient(BaseAPIClient):
        
        Args:
            prompt: 用户提示词
-            system_instruction: 系统指令（可选）
+            model: 模型名称
+            thinking_level: 思考等级（不思考/低/中/高）- 通过动态端点控制，不需要在请求体中传递
            image_data: 图片数据列表，每个元素包含 mime_type 和 data
            video_data: 视频数据，包含 mime_type 和 data
+            document_data: 文档数据，包含 mime_type 和 data
        
        Returns:
            请求体字典
@@ -118,6 +138,15 @@ class GeminiFlashClient(BaseAPIClient):
                }
            })
        
+        # 添加文档部分（如果有）
+        if document_data:
+            parts.append({
+                "inline_data": {
+                    "mime_type": document_data["mime_type"],
+                    "data": document_data["data"]
+                }
+            })
+        
        # 构建请求体
        request_body = {
            "contents": [
@@ -127,12 +156,15 @@ class GeminiFlashClient(BaseAPIClient):
            ]
        }
        
-        # 添加系统指令（如果有）
-        if system_instruction and system_instruction.strip():
-            request_body["system_instruction"] = {
-                "parts": [
-                    {"text": system_instruction}
-                ]
+        # 对于支持 thinkingConfig 的固定端点模型（如 gemini-3-pro-preview）
+        # 通过请求体传递思考等级；动态端点模型（如 gemini-3-flash-preview）
+        # 通过不同 URL 端点控制，无需此字段
+        thinking_level_value = get_flash_model_thinking_level_value(model, thinking_level)
+        if thinking_level_value is not None:
+            request_body["generationConfig"] = {
+                "thinkingConfig": {
+                    "thinkingLevel": thinking_level_value
+                }
            }
        
        return request_body
@@ -197,10 +229,10 @@ class GeminiFlashClient(BaseAPIClient):
        self,
        prompt: str,
        model: str = "gemini-3-flash-preview",
-        thinking_depth: str = "不思考",
-        system_instruction: Optional[str] = None,
+        thinking_level: str = "不思考",
        image_data: Optional[List[Dict[str, str]]] = None,
        video_data: Optional[Dict[str, str]] = None,
+        document_data: Optional[Dict[str, str]] = None,
        session: Optional[aiohttp.ClientSession] = None
    ) -> str:
        """
@@ -209,31 +241,29 @@ class GeminiFlashClient(BaseAPIClient):
        Args:
            prompt: 用户提示词
            model: 模型名称
-            thinking_depth: 思考深度
-            system_instruction: 系统指令
+            thinking_level: 思考等级（不思考/低/中/高）
            image_data: 图片数据列表
            video_data: 视频数据
+            document_data: 文档数据
            session: aiohttp 会话
        
        Returns:
            生成的文本内容
        """
-        endpoint = self.get_endpoint(model=model, thinking_depth=thinking_depth)
+        endpoint = self.get_endpoint(model=model)
        request_body = self.build_request_body(
            prompt=prompt,
-            system_instruction=system_instruction,
+            model=model,
+            thinking_level=thinking_level,
            image_data=image_data,
-            video_data=video_data
+            video_data=video_data,
+            document_data=document_data
        )
        
-        # 根据是否有视频设置超时（视频处理需要更长时间）
-        timeout = 300 if video_data else 180
-        
        response = await self.request_async(
            endpoint, 
            request_body, 
-            session, 
-            timeout=timeout
+            session
        )
        
        return self.parse_response(response)
@@ -242,10 +272,10 @@ class GeminiFlashClient(BaseAPIClient):
        self,
        prompt: str,
        model: str = "gemini-3-flash-preview",
-        thinking_depth: str = "不思考",
-        system_instruction: Optional[str] = None,
+        thinking_level: str = "不思考",
        image_data: Optional[List[Dict[str, str]]] = None,
-        video_data: Optional[Dict[str, str]] = None
+        video_data: Optional[Dict[str, str]] = None,
+        document_data: Optional[Dict[str, str]] = None
    ) -> str:
        """
        同步生成文本（用于 ComfyUI 节点）
@@ -253,10 +283,10 @@ class GeminiFlashClient(BaseAPIClient):
        Args:
            prompt: 用户提示词
            model: 模型名称
-            thinking_depth: 思考深度
-            system_instruction: 系统指令
+            thinking_level: 思考等级（不思考/低/中/高）
            image_data: 图片数据列表
            video_data: 视频数据
+            document_data: 文档数据
        
        Returns:
            生成的文本内容
@@ -264,10 +294,10 @@ class GeminiFlashClient(BaseAPIClient):
        coro = self.generate_async(
            prompt=prompt,
            model=model,
-            thinking_depth=thinking_depth,
-            system_instruction=system_instruction,
+            thinking_level=thinking_level,
            image_data=image_data,
-            video_data=video_data
+            video_data=video_data,
+            document_data=document_data
        )
        
        return self.run_async_in_thread(coro)
@@ -0,0 +1,355 @@
+"""
+Grok Image API 客户端
+支持两个接口：
+  - POST /v1/images/generations  文生图
+  - POST /v1/images/edits        图生图（带参考图）
+
+上游 API 格式与 OpenAI Images API 兼容。
+"""
+
+import asyncio
+import base64
+import json
+import time
+from concurrent.futures import ThreadPoolExecutor
+from io import BytesIO
+from typing import List, Optional
+
+import aiohttp
+import numpy as np
+import torch
+from PIL import Image
+
+from ..utils.config import get_api_key_or_raise, get_base_url_by_route
+from ..utils.image_utils import tensor_to_pil
+
+try:
+    from comfy.model_management import processing_interrupted, InterruptProcessingException
+    _INTERRUPT_AVAILABLE = True
+except ImportError:
+    _INTERRUPT_AVAILABLE = False
+    InterruptProcessingException = RuntimeError
+    processing_interrupted = lambda: False
+
+_ENDPOINT_GENERATIONS = "/v1/images/generations"
+_ENDPOINT_EDITS = "/v1/images/edits"
+
+_MODEL_NAME_MAP = {
+    "Grok Image": "grok-imagine-image",
+    "Grok Image Pro": "grok-imagine-image-quality",
+}
+
+_REQUEST_TIMEOUT = 900
+_MAX_BODY_BYTES = 20 * 1024 * 1024
+
+_MAX_RETRIES = 3
+_RETRY_DELAY = 5
+
+
+class GrokImageClient:
+
+    def __init__(self, route: str = "全球加速"):
+        self.api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        self.base_url = get_base_url_by_route(route)
+
+    def _json_headers(self) -> dict:
+        return {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+
+    def _auth_headers(self) -> dict:
+        return {"Authorization": f"Bearer {self.api_key}"}
+
+    # ── 图像工具 ──────────────────────────────────────────────────────────────
+
+    @staticmethod
+    def _shrink_png_to_limit(png_bytes: bytes, max_bytes: int, label: str = "") -> bytes:
+        if len(png_bytes) <= max_bytes:
+            return png_bytes
+        img = Image.open(BytesIO(png_bytes))
+        w, h = img.size
+        original_size = len(png_bytes)
+        step = 0
+        while len(png_bytes) > max_bytes:
+            scale = 0.894
+            w = max(1, int(w * scale))
+            h = max(1, int(h * scale))
+            img = img.resize((w, h), Image.LANCZOS)
+            buf = BytesIO()
+            img.save(buf, format="PNG")
+            png_bytes = buf.getvalue()
+            step += 1
+        tag = f" ({label})" if label else ""
+        print(
+            f"[o1key Grok Image] 图像{tag}超出 {max_bytes // (1024*1024)}MB 限制，"
+            f"已等比缩放 {step} 次：{original_size // 1024}KB → {len(png_bytes) // 1024}KB "
+            f"（{w}×{h}）"
+        )
+        return png_bytes
+
+    @staticmethod
+    def _pil_list_to_tensor(images: List[Image.Image]) -> torch.Tensor:
+        if not images:
+            placeholder = Image.new("RGB", (512, 512), (128, 128, 128))
+            images = [placeholder]
+        tensors = []
+        for img in images:
+            arr = np.array(img.convert("RGB")).astype(np.float32) / 255.0
+            tensors.append(torch.from_numpy(arr))
+        return torch.stack(tensors, dim=0)
+
+    # ── 中断轮询 ──────────────────────────────────────────────────────────────
+
+    @staticmethod
+    async def _poll_interrupt():
+        while True:
+            await asyncio.sleep(0.5)
+            if _INTERRUPT_AVAILABLE and processing_interrupted():
+                return
+
+    @staticmethod
+    async def _run_with_interrupt(coro):
+        if not _INTERRUPT_AVAILABLE:
+            return await coro
+        request_task = asyncio.ensure_future(coro)
+        interrupt_task = asyncio.ensure_future(GrokImageClient._poll_interrupt())
+        done, pending = await asyncio.wait(
+            [request_task, interrupt_task],
+            return_when=asyncio.FIRST_COMPLETED,
+        )
+        for t in pending:
+            t.cancel()
+            try:
+                await t
+            except (asyncio.CancelledError, Exception):
+                pass
+        if interrupt_task in done and request_task not in done:
+            raise InterruptProcessingException()
+        return request_task.result()
+
+    # ── 响应解析 ──────────────────────────────────────────────────────────────
+
+    async def _parse_response(self, resp_json: dict, session: aiohttp.ClientSession) -> List[Image.Image]:
+        if "error" in resp_json:
+            err = resp_json["error"]
+            msg = (
+                err.get("message") or err.get("msg") or json.dumps(err, ensure_ascii=False)
+                if isinstance(err, dict) else str(err)
+            )
+            raise RuntimeError(f"API 返回错误: {msg}")
+        data_list = resp_json.get("data")
+        if not data_list:
+            raise RuntimeError(f"API 响应中未找到 data 字段")
+        images: List[Image.Image] = []
+        for idx, item in enumerate(data_list):
+            b64 = item.get("b64_json", "")
+            url = item.get("url", "")
+            if b64:
+                img_bytes = base64.b64decode(b64)
+                img = Image.open(BytesIO(img_bytes))
+                images.append(img)
+            elif url and url.startswith("http"):
+                async with session.get(url, allow_redirects=True) as r:
+                    if r.status != 200:
+                        raise RuntimeError(f"图像下载失败 HTTP {r.status}")
+                    img_bytes = await r.read()
+                images.append(Image.open(BytesIO(img_bytes)))
+            else:
+                print(f"[o1key Grok Image] 警告：第 {idx + 1} 条数据无有效图像，已跳过")
+        return images
+
+    # ── 文生图（generations 接口）─────────────────────────────────────────────
+
+    async def _generate_async(
+        self,
+        prompt: str,
+        model: str,
+        aspect_ratio: str,
+        resolution: str,
+        n: int,
+    ) -> List[Image.Image]:
+        api_model = _MODEL_NAME_MAP.get(model, model)
+        body: dict = {
+            "model": api_model,
+            "prompt": prompt,
+            "aspect_ratio": aspect_ratio if aspect_ratio else "auto",
+            "resolution": resolution if resolution else "1k",
+            "response_format": "b64_json",
+        }
+
+        url = f"{self.base_url}{_ENDPOINT_GENERATIONS}"
+        log_body = {k: v for k, v in body.items()}
+        print(f"[o1key Grok Image] 请求 URL: {url}")
+        print(f"[o1key Grok Image] 请求体: {json.dumps(log_body, ensure_ascii=False)}")
+
+        results = []
+        for i in range(n):
+            images = await self._do_request_with_retry(url, body)
+            results.extend(images)
+            if n > 1:
+                print(f"[o1key Grok Image] 第 {i+1}/{n} 张完成")
+        return results
+
+    # ── 图生图（edits 接口）───────────────────────────────────────────────────
+
+    async def _edit_async(
+        self,
+        prompt: str,
+        model: str,
+        aspect_ratio: str,
+        resolution: str,
+        n: int,
+        image_list: List[torch.Tensor],
+    ) -> List[Image.Image]:
+        api_model = _MODEL_NAME_MAP.get(model, model)
+        body: dict = {
+            "model": api_model,
+            "prompt": prompt,
+            "response_format": "b64_json",
+        }
+        if aspect_ratio and aspect_ratio != "auto":
+            body["aspect_ratio"] = aspect_ratio
+        if resolution:
+            body["resolution"] = resolution
+
+        # 参考图转 base64 字符串
+        pil_images = tensor_to_pil(image_list[0])
+        img = pil_images[0]
+        buf = BytesIO()
+        img.save(buf, format="PNG")
+        png_bytes = buf.getvalue()
+        png_bytes = self._shrink_png_to_limit(png_bytes, _MAX_BODY_BYTES // 2)
+        body["image"] = base64.b64encode(png_bytes).decode("utf-8")
+
+        url = f"{self.base_url}{_ENDPOINT_EDITS}"
+        log_body = {k: (v[:50] + "..." if k == "image" and len(v) > 50 else v) for k, v in body.items()}
+        print(f"[o1key Grok Image] 请求 URL: {url}")
+        print(f"[o1key Grok Image] 请求体: {json.dumps(log_body, ensure_ascii=False)}")
+
+        results = []
+        for i in range(n):
+            images = await self._do_request_with_retry(url, body)
+            results.extend(images)
+            if n > 1:
+                print(f"[o1key Grok Image] 第 {i+1}/{n} 张完成")
+        return results
+
+    # ── 带重试的请求 ────────────────────────────────────────────────────────
+
+    async def _do_request_with_retry(self, url: str, body: dict) -> List[Image.Image]:
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        timeout = aiohttp.ClientTimeout(total=_REQUEST_TIMEOUT)
+
+        async def _do_request():
+            async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
+                last_error = None
+                for attempt in range(1, _MAX_RETRIES + 1):
+                    t0 = time.time()
+                    async with session.post(url, json=body, headers=self._json_headers()) as resp:
+                        elapsed = time.time() - t0
+                        text = await resp.text()
+
+                        if resp.status == 429 or resp.status in (502, 503, 504):
+                            last_error = f"HTTP {resp.status}"
+                            print(f"[o1key Grok Image] 重试 {attempt}/{_MAX_RETRIES}（{last_error}）")
+                            await asyncio.sleep(_RETRY_DELAY * attempt)
+                            continue
+
+                        if resp.status == 400 and "high load" in text.lower():
+                            last_error = "high load"
+                            print(f"[o1key Grok Image] 重试 {attempt}/{_MAX_RETRIES}（服务繁忙）")
+                            await asyncio.sleep(_RETRY_DELAY * attempt)
+                            continue
+
+                        if resp.status != 200:
+                            try:
+                                err_json = json.loads(text)
+                                err_obj = err_json.get("error", {})
+                                msg = (
+                                    err_obj.get("message") or err_obj.get("msg") or text
+                                    if isinstance(err_obj, dict) else str(err_obj) or text
+                                )
+                            except Exception:
+                                msg = text
+                            raise RuntimeError(f"请求失败 HTTP {resp.status}: {msg}")
+
+                        try:
+                            resp_json = json.loads(text)
+                        except Exception:
+                            raise RuntimeError(f"响应 JSON 解析失败，原始内容：{text[:500]}")
+
+                    print(f"[o1key Grok Image] API 响应耗时 {elapsed:.1f}s")
+                    return await self._parse_response(resp_json, session)
+
+                raise RuntimeError(f"重试 {_MAX_RETRIES} 次后仍失败: {last_error}")
+
+        return await self._run_with_interrupt(_do_request())
+
+    # ── 同步入口 ──────────────────────────────────────────────────────────────
+
+    def run_sync(
+        self,
+        prompt: str,
+        model: str,
+        aspect_ratio: str,
+        resolution: str,
+        n: int,
+        image_list: Optional[List[torch.Tensor]] = None,
+    ) -> List[Image.Image]:
+        if image_list:
+            coro = self._edit_async(
+                prompt=prompt, model=model, aspect_ratio=aspect_ratio,
+                resolution=resolution, n=n, image_list=image_list,
+            )
+        else:
+            coro = self._generate_async(
+                prompt=prompt, model=model, aspect_ratio=aspect_ratio,
+                resolution=resolution, n=n,
+            )
+
+        def _run():
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            try:
+                return loop.run_until_complete(coro)
+            finally:
+                loop.close()
+
+        with ThreadPoolExecutor(max_workers=1) as executor:
+            future = executor.submit(_run)
+            try:
+                return future.result(timeout=_REQUEST_TIMEOUT + 30)
+            except TimeoutError:
+                raise RuntimeError("Grok Image 请求超时，请检查网络或稍后重试")
+
+    # ── 余额查询 ──────────────────────────────────────────────────────────────
+
+    async def _query_balance_async(self) -> dict:
+        url = f"{self.base_url}/api/usage/token"
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        timeout = aiohttp.ClientTimeout(total=10)
+        async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
+            async with session.get(url, headers=self._auth_headers()) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"余额查询失败 HTTP {resp.status}")
+                return await resp.json()
+
+    def query_balance_sync(self) -> dict:
+        def _run():
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+            try:
+                return loop.run_until_complete(self._query_balance_async())
+            finally:
+                loop.close()
+        with ThreadPoolExecutor(max_workers=1) as executor:
+            return executor.submit(_run).result(timeout=15)
+
+    @staticmethod
+    def format_balance_info(balance_data: dict) -> str:
+        data = balance_data.get("data", {})
+        api_name = data.get("name", "未知")
+        total_available = data.get("total_available", 0)
+        balance_in_dollars = total_available / 500000
+        return f"当前余额：{balance_in_dollars:.2f} | API：{api_name}"
@@ -0,0 +1,483 @@
+"""
+Grok Video API client.
+
+Flow:
+  1. POST /v1/videos
+  2. GET  /v1/videos/{task_id}
+  3. GET  /v1/videos/{task_id}/content, or download a URL from the status body
+"""
+
+import asyncio
+import base64
+import json
+import os
+import re
+import time
+from typing import Any, Callable, Dict, List, Optional
+
+import aiohttp
+
+from .base_client import BaseAPIClient
+from ..utils.config import get_api_base_url, get_api_key_or_raise
+from ..utils.http_error import RETRYABLE_STATUS_CODES, get_friendly_message
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+
+class GrokVideoClient(BaseAPIClient):
+    CREATE_ENDPOINT = "/v1/videos"
+    STATUS_ENDPOINT = "/v1/videos/{task_id}"
+    CONTENT_ENDPOINT = "/v1/videos/{task_id}/content"
+
+    MODEL_OPTIONS = ["grok-imagine-video-1.5-preview", "grok-imagine-1.0-video"]
+    ASPECT_RATIO_OPTIONS = ["1:1", "16:9", "9:16", "4:3", "3:4", "3:2", "2:3"]
+    QUALITY_OPTIONS = ["720p"]
+    MODEL_SECONDS_OPTIONS = {
+        "grok-imagine-1.0-video": [6, 10, 12, 16, 20],
+    }
+    QUALITY_API_MAP = {
+        "720p": "high",
+        "high": "high",
+    }
+
+    SUCCESS_STATUSES = {"complete", "completed", "succeed", "succeeded", "success", "done", "finished"}
+    FAILURE_STATUSES = {"fail", "failed", "failure", "error", "expired", "timeout", "cancelled", "canceled"}
+
+    def __init__(self, base_url: Optional[str] = None):
+        api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        resolved_base_url = (base_url or "").strip() or get_api_base_url()
+        super().__init__(base_url=resolved_base_url.rstrip("/"), api_key=api_key)
+
+    def get_endpoint(self, **kwargs) -> str:
+        return self.CREATE_ENDPOINT
+
+    def build_request_body(self, **kwargs) -> Dict[str, Any]:
+        return self.build_video_body(**kwargs)
+
+    def parse_response(self, response: Dict[str, Any]) -> Any:
+        return response
+
+    @classmethod
+    def build_video_body(
+        cls,
+        prompt: str,
+        model: str,
+        aspect_ratio: str,
+        seconds: int,
+        quality: str = "720p",
+        images: Optional[List[str]] = None,
+    ) -> Dict[str, Any]:
+        prompt = (prompt or "").strip()
+        if not prompt:
+            raise ValueError("提示词不能为空。")
+
+        if model not in cls.MODEL_OPTIONS:
+            raise ValueError(f"模型仅支持: {', '.join(cls.MODEL_OPTIONS)}。")
+
+        if aspect_ratio not in cls.ASPECT_RATIO_OPTIONS:
+            raise ValueError(f"宽高比仅支持: {', '.join(cls.ASPECT_RATIO_OPTIONS)}。")
+
+        try:
+            seconds_value = int(seconds)
+        except (TypeError, ValueError):
+            raise ValueError("秒数必须是整数。") from None
+
+        allowed_seconds = cls.MODEL_SECONDS_OPTIONS.get(model)
+        if allowed_seconds is not None:
+            if seconds_value not in allowed_seconds:
+                raise ValueError(
+                    f"模型 {model} 仅支持秒数: "
+                    f"{', '.join(str(s) for s in allowed_seconds)}。"
+                    "请修改为正确的秒数后再发起请求。"
+                )
+        elif seconds_value < 5 or seconds_value > 15:
+            raise ValueError("秒数仅支持 5 到 15。")
+
+        api_quality = cls.QUALITY_API_MAP.get(str(quality), str(quality))
+        if api_quality != "high":
+            raise ValueError("画质仅支持 720p。")
+
+        body: Dict[str, Any] = {
+            "model": model,
+            "prompt": prompt,
+            "aspect_ratio": aspect_ratio,
+            "seconds": str(seconds_value),
+            "quality": api_quality,
+        }
+
+        image_list = [img for img in (images or []) if img]
+        if image_list:
+            body["images"] = image_list[:3]
+
+        return body
+
+    @staticmethod
+    def _safe_task_filename(task_id: str) -> str:
+        safe = re.sub(r"[^A-Za-z0-9_.-]+", "_", task_id).strip("._")
+        return safe or "grok_video"
+
+    @staticmethod
+    def _mask_body_for_log(body: Dict[str, Any]) -> Dict[str, Any]:
+        log_body = dict(body)
+        images = log_body.get("images")
+        if isinstance(images, list):
+            log_body["images"] = [f"<data-url chars={len(item)}>" for item in images]
+        return log_body
+
+    @staticmethod
+    def _extract_task_id(payload: Dict[str, Any]) -> Optional[str]:
+        sources = [payload]
+        data = payload.get("data")
+        if isinstance(data, dict):
+            sources.append(data)
+
+        for source in sources:
+            for key in ("id", "task_id", "video_id"):
+                value = source.get(key)
+                if value:
+                    return str(value)
+        return None
+
+    @staticmethod
+    def _format_http_error(endpoint: str, status: int, error_text: str, task_id: Optional[str] = None) -> str:
+        message = get_friendly_message(status, error_text)
+        parts = [
+            "Grok Video 请求失败。",
+            f"endpoint: {endpoint}",
+            f"http_status: {status}",
+        ]
+        if task_id:
+            parts.append(f"task_id: {task_id}")
+        if message:
+            parts.append(f"message: {message}")
+        return "\n".join(parts)
+
+    @classmethod
+    def _format_task_failure(cls, task_id: str, payload: Dict[str, Any]) -> str:
+        return "\n".join(
+            [
+                "Grok Video 任务失败。",
+                f"endpoint: {cls.STATUS_ENDPOINT.format(task_id=task_id)}",
+                f"task_id: {task_id}",
+                f"message: {extract_error_message(payload)}",
+            ]
+        )
+
+    async def _request_json_with_retry(
+        self,
+        method: str,
+        endpoint: str,
+        session: aiohttp.ClientSession,
+        task_id: Optional[str] = None,
+        json_body: Optional[Dict[str, Any]] = None,
+        max_retries: int = 3,
+        timeout_seconds: int = 120,
+    ) -> Dict[str, Any]:
+        url = f"{self.base_url}{endpoint}"
+        headers = self.get_headers(use_bearer_token=True)
+        timeout = aiohttp.ClientTimeout(total=timeout_seconds, connect=30, sock_read=timeout_seconds)
+
+        last_status = 0
+        last_text = ""
+
+        for attempt in range(max_retries + 1):
+            check_interrupt()
+            response = None
+            try:
+                response = await run_with_interrupt(
+                    session.request(method, url, json=json_body, headers=headers, timeout=timeout)
+                )
+                text = await run_with_interrupt(response.text())
+                last_status = response.status
+                last_text = text
+
+                if 200 <= response.status < 300:
+                    if not text.strip():
+                        return {}
+                    try:
+                        return json.loads(text)
+                    except Exception:
+                        raise RuntimeError(f"Grok Video 响应 JSON 解析失败，原始内容：{text[:500]}") from None
+
+                if response.status in RETRYABLE_STATUS_CODES and attempt < max_retries:
+                    delay = min(2 ** attempt, 8)
+                    print(
+                        f"Grok Video：{get_friendly_message(response.status)} "
+                        f"{delay}s 后重试 ({attempt + 1}/{max_retries})..."
+                    )
+                    await interruptible_sleep(delay)
+                    continue
+
+                break
+
+            except (aiohttp.ClientError, asyncio.TimeoutError) as e:
+                if attempt < max_retries:
+                    delay = min(2 ** attempt, 8)
+                    print(f"Grok Video：网络错误，{delay}s 后重试 ({attempt + 1}/{max_retries})...")
+                    await interruptible_sleep(delay)
+                    continue
+                raise RuntimeError(f"Grok Video 网络错误: {e}") from None
+
+            finally:
+                if response is not None:
+                    response.release()
+
+        raise RuntimeError(self._format_http_error(endpoint, last_status, last_text, task_id=task_id))
+
+    async def create_video_async(
+        self,
+        body: Dict[str, Any],
+        session: aiohttp.ClientSession,
+    ) -> Dict[str, Any]:
+        print("Grok Video：正在提交任务...")
+        return await self._request_json_with_retry(
+            "POST",
+            self.CREATE_ENDPOINT,
+            session=session,
+            json_body=body,
+            timeout_seconds=180,
+        )
+
+    async def poll_video_status_async(
+        self,
+        task_id: str,
+        session: aiohttp.ClientSession,
+        poll_interval: int = 5,
+        timeout: int = 900,
+        progress_callback: Optional[Callable[[int, str, float], None]] = None,
+    ) -> Dict[str, Any]:
+        endpoint = self.STATUS_ENDPOINT.format(task_id=task_id)
+        start = time.time()
+        interval = max(1, int(poll_interval))
+
+        await interruptible_sleep(interval)
+
+        while True:
+            data = await self._request_json_with_retry(
+                "GET",
+                endpoint,
+                session=session,
+                task_id=task_id,
+                timeout_seconds=60,
+            )
+
+            status = extract_status(data)
+            progress = extract_progress(data)
+            elapsed = time.time() - start
+
+            if progress_callback:
+                progress_callback(progress, status, elapsed)
+
+            if status in self.SUCCESS_STATUSES or is_success_status(status):
+                return data
+
+            if status in self.FAILURE_STATUSES or is_failure_status(status, data):
+                raise RuntimeError(self._format_task_failure(task_id, data))
+
+            if elapsed >= timeout:
+                raise TimeoutError(
+                    "Grok Video 任务轮询超时；任务未被标记为失败，可用 task_id 继续查询。\n"
+                    f"endpoint: {endpoint}\n"
+                    f"task_id: {task_id}\n"
+                    f"status: {status or 'unknown'}\n"
+                    f"timeout: {timeout}s"
+                )
+
+            await interruptible_sleep(min(interval, max(0.0, timeout - elapsed)))
+
+    async def _download_url_to_file(
+        self,
+        url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+        max_retries: int = 3,
+    ) -> str:
+        timeout = aiohttp.ClientTimeout(total=900, connect=30, sock_read=900)
+        last_status = 0
+        last_text = ""
+        headers = None
+        resolved_url = url
+
+        if url.startswith("data:"):
+            if "," not in url:
+                raise RuntimeError("Grok Video 下载失败：data URL 格式无效。")
+            _, b64_data = url.split(",", 1)
+            os.makedirs(os.path.dirname(save_path), exist_ok=True)
+            with open(save_path, "wb") as f:
+                f.write(base64.b64decode(b64_data))
+            if not os.path.isfile(save_path) or os.path.getsize(save_path) <= 0:
+                raise RuntimeError("Grok Video 下载失败：保存后的文件为空。")
+            return save_path
+
+        if url.startswith("/"):
+            resolved_url = f"{self.base_url}{url}"
+            headers = self.get_headers(use_bearer_token=True)
+
+        for attempt in range(max_retries + 1):
+            check_interrupt()
+            async with session.get(
+                resolved_url,
+                headers=headers,
+                timeout=timeout,
+                allow_redirects=True,
+            ) as response:
+                if 200 <= response.status < 300:
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in response.content.iter_chunked(1024 * 1024):
+                            check_interrupt()
+                            if chunk:
+                                f.write(chunk)
+
+                    if not os.path.isfile(save_path) or os.path.getsize(save_path) <= 0:
+                        raise RuntimeError("Grok Video 下载失败：保存后的文件为空。")
+                    return save_path
+
+                last_status = response.status
+                last_text = await response.text()
+                if response.status not in RETRYABLE_STATUS_CODES or attempt >= max_retries:
+                    break
+
+            delay = min(2 ** attempt, 8)
+            print(f"Grok Video：下载重试 {attempt + 1}/{max_retries}，{delay}s 后继续...")
+            await interruptible_sleep(delay)
+
+        raise RuntimeError(self._format_http_error("download_url", last_status, last_text))
+
+    async def download_video_async(
+        self,
+        task_id: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+    ) -> str:
+        endpoint = self.CONTENT_ENDPOINT.format(task_id=task_id)
+        url = f"{self.base_url}{endpoint}"
+        headers = self.get_headers(use_bearer_token=True)
+        timeout = aiohttp.ClientTimeout(total=900, connect=30, sock_read=900)
+
+        last_status = 0
+        last_text = ""
+
+        for attempt in range(4):
+            check_interrupt()
+            async with session.get(url, headers=headers, timeout=timeout, allow_redirects=True) as response:
+                if 200 <= response.status < 300:
+                    content_type = response.headers.get("Content-Type", "").lower()
+                    if "application/json" in content_type:
+                        data = await response.json(content_type=None)
+                        download_url = extract_video_url(data)
+                        if not download_url:
+                            raise RuntimeError(
+                                "Grok Video 下载失败：content 响应为 JSON，但未包含视频 URL。\n"
+                                f"endpoint: {endpoint}\n"
+                                f"task_id: {task_id}"
+                            )
+                        return await self._download_url_to_file(download_url, save_path, session)
+
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in response.content.iter_chunked(1024 * 1024):
+                            check_interrupt()
+                            if chunk:
+                                f.write(chunk)
+
+                    if not os.path.isfile(save_path) or os.path.getsize(save_path) <= 0:
+                        raise RuntimeError(
+                            "Grok Video 下载失败：保存后的文件为空。\n"
+                            f"endpoint: {endpoint}\n"
+                            f"task_id: {task_id}"
+                        )
+                    return save_path
+
+                last_status = response.status
+                last_text = await response.text()
+                if response.status not in RETRYABLE_STATUS_CODES or attempt >= 3:
+                    break
+
+            delay = min(2 ** attempt, 8)
+            print(f"Grok Video：content 下载重试 {attempt + 1}/3，{delay}s 后继续...")
+            await interruptible_sleep(delay)
+
+        raise RuntimeError(self._format_http_error(endpoint, last_status, last_text, task_id=task_id))
+
+    def generate_video_sync(
+        self,
+        prompt: str,
+        model: str,
+        aspect_ratio: str,
+        seconds: int,
+        quality: str,
+        images: Optional[List[str]],
+        output_dir: Optional[str] = None,
+        save_path: Optional[str] = None,
+        poll_interval: int = 5,
+        timeout: int = 900,
+        progress_callback: Optional[Callable[[int, str, float], None]] = None,
+    ) -> Dict[str, Any]:
+        async def _run():
+            async with self._make_session() as session:
+                body = self.build_video_body(
+                    prompt=prompt,
+                    model=model,
+                    aspect_ratio=aspect_ratio,
+                    seconds=seconds,
+                    quality=quality,
+                    images=images,
+                )
+
+                create_response = await self.create_video_async(body, session)
+                task_id = self._extract_task_id(create_response) or ""
+                if not task_id:
+                    raise RuntimeError(
+                        "Grok Video 未返回任务 ID。\n"
+                        f"endpoint: {self.CREATE_ENDPOINT}\n"
+                        f"response: {json.dumps(create_response, ensure_ascii=False)[:1200]}"
+                    )
+
+                print(f"Grok Video：任务已提交，任务ID：{task_id}")
+                print("Grok Video：视频生成中...")
+                status_response = await self.poll_video_status_async(
+                    task_id=task_id,
+                    session=session,
+                    poll_interval=poll_interval,
+                    timeout=timeout,
+                    progress_callback=progress_callback,
+                )
+
+                video_url = extract_video_url(status_response)
+                print("Grok Video：视频生成完成，正在下载...")
+                if save_path is None:
+                    resolved_output_dir = output_dir or os.getcwd()
+                    os.makedirs(resolved_output_dir, exist_ok=True)
+                    target_path = os.path.join(
+                        resolved_output_dir,
+                        f"{self._safe_task_filename(task_id)}.mp4",
+                    )
+                else:
+                    target_path = save_path
+
+                if video_url:
+                    video_path = await self._download_url_to_file(video_url, target_path, session)
+                else:
+                    video_path = await self.download_video_async(task_id, target_path, session)
+
+                return {
+                    "task_id": task_id,
+                    "status": extract_status(status_response),
+                    "video_path": video_path,
+                    "raw_json": {
+                        "create": create_response,
+                        "status": status_response,
+                    },
+                }
+
+        return self.run_async_in_thread(_run())
@@ -0,0 +1,285 @@
+"""
+Kling 视频生成 API 客户端
+"""
+
+import asyncio
+import json
+import os
+from typing import Any, Callable, Dict, Optional
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+
+class KlingClient:
+    """Kling 视频生成客户端"""
+
+    ENDPOINTS = {
+        "image2video":    "/kling/v1/videos/image2video",
+        "text2video":     "/kling/v1/videos/text2video",
+        "motion_control": "/kling/v1/videos/motion-control",
+    }
+
+    # new API 三段式端点（动作控制走这里）
+    NEW_API_CREATE   = "/v1/videos"
+    NEW_API_STATUS   = "/v1/videos/{video_id}"
+    NEW_API_CONTENT  = "/v1/videos/{video_id}/content"
+
+    POLL_INITIAL_INTERVAL = 3
+    POLL_MAX_INTERVAL = 15
+
+    def __init__(self):
+        self.api_key = get_api_key_or_raise()
+        self.base_url = get_api_base_url()
+
+    def _headers(self) -> Dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+
+    # ── 提交任务 ──────────────────────────────────────────────────────
+
+    async def create_video_async(
+        self,
+        endpoint_type: str,
+        body: Dict[str, Any],
+        session: aiohttp.ClientSession,
+    ) -> Dict[str, Any]:
+        url = f"{self.base_url}{self.ENDPOINTS[endpoint_type]}"
+
+        check_interrupt()
+        resp = await run_with_interrupt(async_request_with_retry(
+            session, "POST", url, json=body, headers=self._headers(), prefix="Kling 提交: "
+        ))
+        check_interrupt()
+        text = await resp.text()
+        return json.loads(text)
+
+    # ── 轮询状态 ──────────────────────────────────────────────────────
+
+    async def poll_status_async(
+        self,
+        task_id: str,
+        endpoint_type: str,
+        session: aiohttp.ClientSession,
+        on_progress: Optional[Callable[[int], None]] = None,
+    ) -> Dict[str, Any]:
+        url = f"{self.base_url}{self.ENDPOINTS[endpoint_type]}/{task_id}"
+        interval = self.POLL_INITIAL_INTERVAL
+
+        while True:
+            check_interrupt()
+            async with session.get(url, headers=self._headers()) as resp:
+                text = await resp.text()
+                if resp.status != 200:
+                    raise RuntimeError(f"状态查询失败 ({resp.status}): {text}")
+                result = json.loads(text)
+
+            data = result.get("data", {})
+            inner_data = data.get("data", {}) if isinstance(data, dict) else {}
+            status = extract_status(result)
+
+            progress_pct = extract_progress(result)
+
+            print(f"[视频生成] 生成中 {progress_pct}%")
+
+            if on_progress:
+                on_progress(progress_pct)
+
+            if is_success_status(status):
+                return result
+            elif is_failure_status(status, result):
+                error_msg = extract_error_message(result)
+                raise RuntimeError(f"生成失败：{error_msg}")
+
+            await interruptible_sleep(interval)
+            interval = min(interval * 1.5, self.POLL_MAX_INTERVAL)
+
+    # ── 下载视频 ──────────────────────────────────────────────────────
+
+    async def download_video_async(
+        self,
+        video_url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+    ) -> str:
+        print("[视频生成] 下载视频...")
+        check_interrupt()
+        async with session.get(video_url, allow_redirects=True) as resp:
+            if resp.status != 200:
+                raise RuntimeError(f"视频下载失败 ({resp.status})")
+            os.makedirs(os.path.dirname(save_path), exist_ok=True)
+            with open(save_path, "wb") as f:
+                async for chunk in resp.content.iter_chunked(8192):
+                    check_interrupt()
+                    f.write(chunk)
+        return save_path
+
+    # ── 异步入口（供节点调用）────────────────────────────────────────
+
+    async def generate_async(
+        self,
+        endpoint_type: str,
+        body: Dict[str, Any],
+        save_path: str,
+        on_stage: Optional[Callable[[str], None]] = None,
+        on_progress: Optional[Callable[[int], None]] = None,
+    ) -> str:
+        """提交 → 轮询 → 下载，返回本地文件路径"""
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+            if on_stage:
+                on_stage("submitting")
+
+            result = await self.create_video_async(endpoint_type, body, session)
+            # 提交响应结构：result.data.task_id
+            task_id = result.get("task_id") or result.get("data", {}).get("task_id")
+            if not task_id:
+                raise RuntimeError(f"API 未返回任务 ID，响应：{result}")
+            if on_stage:
+                on_stage(f"submitted:{task_id}")
+
+            if on_stage:
+                on_stage("polling")
+            final = await self.poll_status_async(
+                task_id, endpoint_type, session, on_progress=on_progress
+            )
+
+            # 兼容多种URL路径
+            # 响应结构：result.data.result_url 或 result.data.data.task_result.videos[0].url
+            data = final.get("data", {})
+            inner_data = data.get("data", {}) if isinstance(data, dict) else {}
+            video_url = (
+                data.get("result_url") or
+                final.get("url") or
+                final.get("video_url") or
+                (inner_data.get("task_result", {}).get("videos", [{}])[0].get("url")
+                 if inner_data.get("task_result", {}).get("videos") else None)
+            )
+            if not video_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{final}")
+
+            if on_stage:
+                on_stage("downloading")
+            path = await self.download_video_async(video_url, save_path, session)
+
+            if on_stage:
+                on_stage("done")
+            return path
+
+    # ── 动作控制：走 new API 三段式流程 ──────────────────────────────
+
+    async def motion_control_async(
+        self,
+        body: Dict[str, Any],
+        save_path: str,
+        on_stage: Optional[Callable[[str], None]] = None,
+        on_progress: Optional[Callable[[int], None]] = None,
+    ) -> str:
+        """
+        动作控制专用入口：
+          POST /v1/videos  →  GET /v1/videos/{id}  →  GET /v1/videos/{id}/content
+        body 字段与 Kling 官方动作控制接口一致（image_url/video_url/prompt/...）。
+        """
+        headers = {"Authorization": f"Bearer {self.api_key}",
+                   "Content-Type": "application/json"}
+        interval = self.POLL_INITIAL_INTERVAL
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交
+            check_interrupt()
+            if on_stage:
+                on_stage("submitting")
+            create_url = f"{self.base_url}{self.NEW_API_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url, json=body, headers=headers, prefix="Kling 动作控制提交: "
+            ))
+            check_interrupt()
+            text = await resp.text()
+            create_resp = json.loads(text)
+
+            video_id = create_resp.get("id")
+            if not video_id:
+                raise RuntimeError(f"API 未返回视频 ID，响应：{create_resp}")
+            if on_stage:
+                on_stage(f"submitted:{video_id}")
+
+            # 2. 轮询
+            status_url = f"{self.base_url}{self.NEW_API_STATUS.format(video_id=video_id)}"
+            while True:
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    text = await resp.text()
+                    if resp.status != 200:
+                        try:
+                            err = json.loads(text)
+                            msg = err.get("error", {}).get("message") or err.get("message") or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                    status_resp = json.loads(text)
+
+                status = extract_status(status_resp)
+                progress_pct = extract_progress(status_resp)
+
+                print(f"[动作控制] 生成中 {progress_pct}%")
+                if on_progress:
+                    on_progress(progress_pct)
+
+                if is_success_status(status):
+                    break
+                if is_failure_status(status, status_resp):
+                    error_msg = extract_error_message(status_resp)
+                    raise RuntimeError(f"动作控制生成失败：{error_msg}")
+
+                await interruptible_sleep(interval)
+                interval = min(interval * 1.5, self.POLL_MAX_INTERVAL)
+
+            # 3. 下载
+            check_interrupt()
+            if on_stage:
+                on_stage("downloading")
+            content_url = f"{self.base_url}{self.NEW_API_CONTENT.format(video_id=video_id)}"
+            async with session.get(content_url, headers=headers,
+                                   allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                content_type = resp.headers.get("Content-Type", "")
+                if "application/json" in content_type:
+                    data = await resp.json()
+                    download_url = data.get("url") or data.get("download_url")
+                    if not download_url:
+                        raise RuntimeError("视频下载失败：响应中未找到下载链接")
+                    async with session.get(download_url) as dl_resp:
+                        if dl_resp.status != 200:
+                            raise RuntimeError(f"从下载链接获取视频失败 ({dl_resp.status})")
+                        os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                        with open(save_path, "wb") as f:
+                            async for chunk in dl_resp.content.iter_chunked(8192):
+                                check_interrupt()
+                                f.write(chunk)
+                else:
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in resp.content.iter_chunked(8192):
+                            check_interrupt()
+                            f.write(chunk)
+
+            if on_stage:
+                on_stage("done")
+            return save_path
@@ -0,0 +1,542 @@
+"""
+new-api Veo 3.1 video client.
+
+Implements the OpenAI-compatible /v1/videos task flow:
+submit, poll, and stream-download video content.
+"""
+
+import asyncio
+import json
+import os
+import re
+import time
+from typing import Any, Callable, Dict, Optional
+
+import aiohttp
+
+from .base_client import BaseAPIClient
+from ..utils.config import get_api_base_url, get_api_key_or_raise
+
+
+class NewAPIVeoClient(BaseAPIClient):
+    CREATE_ENDPOINT = "/v1/videos"
+    STATUS_ENDPOINT = "/v1/videos/{task_id}"
+    CONTENT_ENDPOINT = "/v1/videos/{task_id}/content"
+
+    RETRYABLE_STATUS_CODES = {408, 409, 425, 429, 500, 502, 503, 504}
+    COMPLETED_STATUSES = {"completed", "succeeded", "success", "done"}
+    FAILED_STATUSES = {"failed", "error", "cancelled", "canceled"}
+
+    def __init__(
+        self,
+        base_url: Optional[str] = None,
+    ):
+        api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        resolved_base_url = (base_url or "").strip() or get_api_base_url()
+        super().__init__(base_url=resolved_base_url.rstrip("/"), api_key=api_key)
+
+    def get_endpoint(self, **kwargs) -> str:
+        return self.CREATE_ENDPOINT
+
+    def build_request_body(self, **kwargs) -> Dict[str, Any]:
+        return self._build_video_body(**kwargs)
+
+    def parse_response(self, response: Dict[str, Any]) -> Any:
+        return response
+
+    @staticmethod
+    def _build_video_body(
+        prompt: str,
+        model: str,
+        duration: int,
+        aspect_ratio: str,
+        resolution: str,
+        negative_prompt: str = "",
+        generate_audio: bool = True,
+    ) -> Dict[str, Any]:
+        metadata: Dict[str, Any] = {
+            "aspectRatio": aspect_ratio,
+            "resolution": resolution,
+            "generateAudio": bool(generate_audio),
+        }
+
+        negative_prompt = (negative_prompt or "").strip()
+        if negative_prompt:
+            metadata["negativePrompt"] = negative_prompt
+
+        body: Dict[str, Any] = {
+            "model": model,
+            "prompt": prompt,
+            "duration": int(duration),
+            "metadata": metadata,
+        }
+
+        return body
+
+    @staticmethod
+    def _print_request_body(body: Dict[str, Any], image_bytes: Optional[bytes] = None) -> None:
+        log_body = dict(body)
+        if image_bytes is not None:
+            log_body["input_reference"] = f"<PNG bytes: {len(image_bytes)}>"
+        print(
+            "NewAPI Veo request body:\n"
+            f"{json.dumps(log_body, ensure_ascii=False, indent=2)}"
+        )
+
+    @staticmethod
+    def _safe_task_filename(task_id: str) -> str:
+        safe = re.sub(r"[^A-Za-z0-9_.-]+", "_", task_id).strip("._")
+        return safe or "newapi_veo"
+
+    @staticmethod
+    def _extract_task_id(data: Dict[str, Any]) -> Optional[str]:
+        for key in ("id", "task_id", "video_id"):
+            value = data.get(key)
+            if value:
+                return str(value)
+
+        nested = data.get("data")
+        if isinstance(nested, dict):
+            for key in ("id", "task_id", "video_id"):
+                value = nested.get(key)
+                if value:
+                    return str(value)
+        return None
+
+    @staticmethod
+    def _extract_status(data: Dict[str, Any]) -> str:
+        for key in ("status", "state", "task_status"):
+            value = data.get(key)
+            if value:
+                return str(value).lower()
+
+        nested = data.get("data")
+        if isinstance(nested, dict):
+            for key in ("status", "state", "task_status"):
+                value = nested.get(key)
+                if value:
+                    return str(value).lower()
+        return "unknown"
+
+    @staticmethod
+    def _extract_progress(data: Dict[str, Any]) -> int:
+        progress = data.get("progress")
+        if progress is None and isinstance(data.get("data"), dict):
+            progress = data["data"].get("progress")
+
+        if isinstance(progress, str):
+            progress = progress.rstrip("%").strip()
+            try:
+                return int(float(progress))
+            except ValueError:
+                return 0
+        if isinstance(progress, (int, float)):
+            return int(progress)
+        return 0
+
+    @classmethod
+    def _format_http_error(
+        cls,
+        endpoint: str,
+        status: int,
+        error_text: str,
+        task_id: Optional[str] = None,
+    ) -> str:
+        code = ""
+        message = error_text
+        try:
+            payload = json.loads(error_text)
+            error = payload.get("error", payload)
+            if isinstance(error, dict):
+                code = str(error.get("code") or error.get("type") or "")
+                message = str(error.get("message") or payload.get("message") or error_text)
+            elif error is not None:
+                message = str(error)
+        except Exception:
+            pass
+
+        message = (message or "").strip()
+        if len(message) > 1200:
+            message = message[:1200] + "...(truncated)"
+
+        if status in (401, 403):
+            hint = "凭证或分组权限问题，请检查 new-api token、模型分组或渠道权限。"
+        elif status == 429:
+            hint = "频率或额度限制，请稍后重试或检查 new-api 额度。"
+        elif status in (502, 503, 504):
+            hint = "上游服务暂时不可用或超时，请稍后用 task_id 继续查询。"
+        elif status == 400:
+            hint = "请求参数错误，请检查 model、duration、metadata 和图片输入。"
+        else:
+            hint = "new-api 视频请求失败。"
+
+        parts = [
+            hint,
+            f"endpoint: {endpoint}",
+            f"http_status: {status}",
+        ]
+        if task_id:
+            parts.append(f"task_id: {task_id}")
+        if code:
+            parts.append(f"error_code: {code}")
+        if message:
+            parts.append(f"message: {message}")
+        return "\n".join(parts)
+
+    @classmethod
+    def _format_task_failure(cls, task_id: str, data: Dict[str, Any]) -> str:
+        error = data.get("error")
+        if error is None and isinstance(data.get("data"), dict):
+            error = data["data"].get("error")
+
+        if isinstance(error, dict):
+            code = error.get("code") or error.get("type") or ""
+            message = error.get("message") or json.dumps(error, ensure_ascii=False)
+        else:
+            code = ""
+            message = str(error or "未知错误")
+
+        return "\n".join(
+            [
+                "Veo 视频任务失败。",
+                f"endpoint: {cls.STATUS_ENDPOINT.format(task_id=task_id)}",
+                f"task_id: {task_id}",
+                f"error_code: {code}",
+                f"message: {message}",
+            ]
+        )
+
+    async def create_video_async(
+        self,
+        prompt: str,
+        model: str,
+        duration: int,
+        aspect_ratio: str,
+        resolution: str,
+        negative_prompt: str = "",
+        generate_audio: bool = True,
+        image_bytes: Optional[bytes] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        body = self._build_video_body(
+            prompt=prompt,
+            model=model,
+            duration=duration,
+            aspect_ratio=aspect_ratio,
+            resolution=resolution,
+            negative_prompt=negative_prompt,
+            generate_audio=generate_audio,
+        )
+
+        url = f"{self.base_url}{self.CREATE_ENDPOINT}"
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            timeout = aiohttp.ClientTimeout(total=120, connect=30, sock_read=120)
+            headers = {"Authorization": f"Bearer {self.api_key}"}
+
+            if image_bytes is not None:
+                if len(image_bytes) > self.max_request_size:
+                    raise ValueError(
+                        f"输入图片过大，超过 {self.max_request_size / 1024 / 1024:.0f}MB 限制"
+                    )
+
+                self._print_request_body(body, image_bytes=image_bytes)
+                form = aiohttp.FormData()
+                form.add_field("model", body["model"])
+                form.add_field("prompt", body["prompt"])
+                form.add_field("duration", str(body["duration"]))
+                form.add_field("metadata", json.dumps(body["metadata"], ensure_ascii=False))
+                form.add_field(
+                    "input_reference",
+                    image_bytes,
+                    filename="input_reference.png",
+                    content_type="image/png",
+                )
+                request_kwargs = {"data": form, "headers": headers}
+                print(
+                    "NewAPI Veo: POST /v1/videos multipart "
+                    f"| model={model} | duration={duration}s | {resolution} {aspect_ratio}"
+                )
+            else:
+                self._print_request_body(body)
+                headers["Content-Type"] = "application/json"
+                request_kwargs = {"json": body, "headers": headers}
+                print(
+                    "NewAPI Veo: POST /v1/videos json "
+                    f"| model={model} | duration={duration}s | {resolution} {aspect_ratio}"
+                )
+
+            async with session.post(url, timeout=timeout, **request_kwargs) as response:
+                if response.status >= 300:
+                    error_text = await response.text()
+                    raise RuntimeError(
+                        self._format_http_error(
+                            self.CREATE_ENDPOINT,
+                            response.status,
+                            error_text,
+                        )
+                    )
+                return await response.json()
+        finally:
+            if close_session:
+                await session.close()
+
+    async def _get_json_with_retry(
+        self,
+        endpoint: str,
+        session: aiohttp.ClientSession,
+        task_id: Optional[str] = None,
+        max_retries: int = 3,
+    ) -> Dict[str, Any]:
+        url = f"{self.base_url}{endpoint}"
+        headers = self.get_headers(use_bearer_token=True)
+        timeout = aiohttp.ClientTimeout(total=60, connect=30, sock_read=60)
+
+        last_error = ""
+        last_status = 0
+        for attempt in range(max_retries + 1):
+            async with session.get(url, headers=headers, timeout=timeout) as response:
+                if response.status < 300:
+                    return await response.json()
+
+                last_status = response.status
+                last_error = await response.text()
+                if response.status not in self.RETRYABLE_STATUS_CODES or attempt >= max_retries:
+                    break
+
+            await asyncio.sleep(min(2 ** attempt, 8))
+
+        raise RuntimeError(
+            self._format_http_error(endpoint, last_status, last_error, task_id=task_id)
+        )
+
+    async def poll_video_status_async(
+        self,
+        task_id: str,
+        poll_interval: int = 5,
+        timeout: int = 900,
+        progress_callback: Optional[Callable[[int, str, float], None]] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        endpoint = self.STATUS_ENDPOINT.format(task_id=task_id)
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        start = time.time()
+        try:
+            poll_interval = max(1, int(poll_interval))
+            await asyncio.sleep(poll_interval)
+
+            while True:
+                data = await self._get_json_with_retry(endpoint, session, task_id=task_id)
+                status = self._extract_status(data)
+                elapsed = time.time() - start
+                progress = self._extract_progress(data)
+
+                if status == "unknown":
+                    print(
+                        "NewAPI Veo status response did not include a recognized status field:\n"
+                        f"{json.dumps(data, ensure_ascii=False, indent=2)[:1200]}"
+                    )
+
+                if progress_callback:
+                    progress_callback(progress, status, elapsed)
+
+                if status in self.COMPLETED_STATUSES:
+                    return data
+
+                if status in self.FAILED_STATUSES:
+                    raise RuntimeError(self._format_task_failure(task_id, data))
+
+                if elapsed >= timeout:
+                    raise TimeoutError(
+                        "Veo 视频任务轮询超时；任务未被标记为失败，可用 task_id 继续查询。\n"
+                        f"endpoint: {endpoint}\n"
+                        f"task_id: {task_id}\n"
+                        f"status: {status}\n"
+                        f"timeout: {timeout}s"
+                    )
+
+                remaining = max(0.0, timeout - elapsed)
+                await asyncio.sleep(min(poll_interval, remaining))
+        finally:
+            if close_session:
+                await session.close()
+
+    async def _download_url_to_file(
+        self,
+        url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+        max_retries: int = 3,
+    ) -> None:
+        timeout = aiohttp.ClientTimeout(total=900, connect=30, sock_read=900)
+        last_status = 0
+        last_error = ""
+
+        for attempt in range(max_retries + 1):
+            async with session.get(url, timeout=timeout, allow_redirects=True) as response:
+                if response.status < 300:
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in response.content.iter_chunked(1024 * 1024):
+                            if chunk:
+                                f.write(chunk)
+                    return
+
+                last_status = response.status
+                last_error = await response.text()
+                if response.status not in self.RETRYABLE_STATUS_CODES or attempt >= max_retries:
+                    break
+
+            await asyncio.sleep(min(2 ** attempt, 8))
+
+        raise RuntimeError(
+            self._format_http_error("download_url", last_status, last_error)
+        )
+
+    async def download_video_async(
+        self,
+        task_id: str,
+        save_path: str,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> str:
+        endpoint = self.CONTENT_ENDPOINT.format(task_id=task_id)
+        url = f"{self.base_url}{endpoint}"
+        headers = self.get_headers(use_bearer_token=True)
+        timeout = aiohttp.ClientTimeout(total=900, connect=30, sock_read=900)
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            last_status = 0
+            last_error = ""
+            for attempt in range(4):
+                async with session.get(url, headers=headers, timeout=timeout, allow_redirects=True) as response:
+                    if response.status < 300:
+                        content_type = response.headers.get("Content-Type", "")
+                        if "application/json" in content_type.lower():
+                            data = await response.json()
+                            nested = data.get("data") if isinstance(data.get("data"), dict) else {}
+                            download_url = (
+                                data.get("url")
+                                or data.get("download_url")
+                                or nested.get("url")
+                                or nested.get("download_url")
+                            )
+                            if not download_url:
+                                raise RuntimeError(
+                                    "视频下载失败: content 响应为 JSON，但未包含 url/download_url。\n"
+                                    f"endpoint: {endpoint}\n"
+                                    f"task_id: {task_id}"
+                                )
+                            await self._download_url_to_file(download_url, save_path, session)
+                        else:
+                            os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                            with open(save_path, "wb") as f:
+                                async for chunk in response.content.iter_chunked(1024 * 1024):
+                                    if chunk:
+                                        f.write(chunk)
+
+                        if not os.path.isfile(save_path) or os.path.getsize(save_path) <= 0:
+                            raise RuntimeError(
+                                "视频下载失败: 保存后的文件为空。\n"
+                                f"endpoint: {endpoint}\n"
+                                f"task_id: {task_id}"
+                            )
+                        return save_path
+
+                    last_status = response.status
+                    last_error = await response.text()
+                    if response.status not in self.RETRYABLE_STATUS_CODES or attempt >= 3:
+                        break
+
+                await asyncio.sleep(min(2 ** attempt, 8))
+
+            raise RuntimeError(
+                self._format_http_error(endpoint, last_status, last_error, task_id=task_id)
+            )
+        finally:
+            if close_session:
+                await session.close()
+
+    def generate_video_sync(
+        self,
+        prompt: str,
+        model: str,
+        duration: int,
+        aspect_ratio: str,
+        resolution: str,
+        output_dir: str,
+        negative_prompt: str = "",
+        generate_audio: bool = True,
+        image_bytes: Optional[bytes] = None,
+        poll_interval: int = 5,
+        timeout: int = 900,
+        reuse_task_id: str = "",
+        progress_callback: Optional[Callable[[int, str, float], None]] = None,
+    ) -> Dict[str, Any]:
+        async def _run():
+            async with self._make_session() as session:
+                create_response: Dict[str, Any] = {}
+                task_id = (reuse_task_id or "").strip()
+                if task_id:
+                    print(f"NewAPI Veo: reuse task_id={task_id}")
+                else:
+                    create_response = await self.create_video_async(
+                        prompt=prompt,
+                        model=model,
+                        duration=duration,
+                        aspect_ratio=aspect_ratio,
+                        resolution=resolution,
+                        negative_prompt=negative_prompt,
+                        generate_audio=generate_audio,
+                        image_bytes=image_bytes,
+                        session=session,
+                    )
+                    task_id = self._extract_task_id(create_response) or ""
+                    if not task_id:
+                        raise RuntimeError(
+                            "new-api 未返回视频任务 ID。\n"
+                            f"endpoint: {self.CREATE_ENDPOINT}\n"
+                            f"response: {json.dumps(create_response, ensure_ascii=False)[:1200]}"
+                        )
+
+                status_response = await self.poll_video_status_async(
+                    task_id=task_id,
+                    poll_interval=poll_interval,
+                    timeout=timeout,
+                    progress_callback=progress_callback,
+                    session=session,
+                )
+                status = self._extract_status(status_response)
+
+                os.makedirs(output_dir, exist_ok=True)
+                filename = f"{self._safe_task_filename(task_id)}.mp4"
+                save_path = os.path.join(output_dir, filename)
+                video_path = await self.download_video_async(
+                    task_id=task_id,
+                    save_path=save_path,
+                    session=session,
+                )
+
+                return {
+                    "task_id": task_id,
+                    "status": status,
+                    "video_path": video_path,
+                    "raw_json": {
+                        "create": create_response,
+                        "status": status_response,
+                    },
+                }
+
+        return self.run_async_in_thread(_run())
@@ -0,0 +1,784 @@
+"""
+OpenAI 兼容 API 客户端
+端点固定为 /v1/chat/completions，模型名放入请求体 model 字段
+"""
+
+import re
+import time
+from io import BytesIO
+from typing import Any, Callable, Dict, List, Optional
+
+import aiohttp
+from PIL import Image
+
+from ..utils.image_utils import encode_image_to_base64, decode_base64_to_pil
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from .base_client import BaseAPIClient
+
+
+# 固定端点
+_ENDPOINT = "/v1/chat/completions"
+
+
+class OpenAIAPIClient(BaseAPIClient):
+    """
+    OpenAI 兼容格式的图像生成客户端
+
+    与 GeminiAPIClient 的主要区别：
+    - 端点固定为 /v1/chat/completions（不再动态拼模型名到 URL）
+    - 解析后的模型字符串放入请求体的 model 字段
+    - 请求体采用 messages 数组格式，图片以 data URI 内联
+    - 顶层追加 modalities 和 image_config 字段
+    - 响应解析对应 choices[0].message.content 结构
+    """
+
+    def __init__(self, api_key: Optional[str] = None):
+        if api_key is None:
+            api_key = get_api_key_or_raise("O1KEY_API_KEY")
+
+        super().__init__(
+            base_url=get_api_base_url(),
+            api_key=api_key,
+            max_request_size=100 * 1024 * 1024
+        )
+
+    # ------------------------------------------------------------------ #
+    #  模型名解析                                                           #
+    #  原 GeminiAPIClient.get_endpoint() 里动态拼 URL 的逻辑               #
+    #  现在改为：同样的输入 → 返回纯模型名字符串，放进请求体                    #
+    # ------------------------------------------------------------------ #
+
+    def resolve_model_name(self, model: str, resolution: str) -> str:
+        """
+        将「节点选中的模型 ID + 分辨率」解析为实际请求所用的模型名称。
+
+        对应关系与原 GeminiAPIClient.get_endpoint() 完全一致，
+        只是把拼在 URL 路径里的模型段提取出来单独返回。
+
+        Args:
+            model:      节点下拉框中的模型 ID，如 "nano-banana-pro-次卡"
+            resolution: 分辨率字符串，如 "1K" / "2K" / "4K" / "512"
+
+        Returns:
+            实际模型名，如 "nano-banana-pro-2k"
+        """
+        # ── 动态端点模型 ──────────────────────────────────────────────────
+        if model == "nano-banana-pro-次卡":
+            if resolution == "1K":
+                return "nano-banana-pro"
+            elif resolution == "4K":
+                return "nano-banana-pro-4k"
+            else:  # 2K（默认）
+                return "nano-banana-pro-2k"
+
+        elif model == "nano-banana-pro-官方计费":
+            if resolution == "1K":
+                return "nano-banana-pro-1k-official"
+            elif resolution == "4K":
+                return "nano-banana-pro-4k-official"
+            else:  # 2K（默认）
+                return "nano-banana-pro-2k-official"
+
+        elif model == "nano-banana-2-官方计费":
+            if resolution == "512":
+                return "nano-banana-2-0.5k-official"
+            elif resolution == "1K":
+                return "nano-banana-2-1k-official"
+            elif resolution == "4K":
+                return "nano-banana-2-4k-official"
+            else:  # 2K（默认）
+                return "nano-banana-2-2k-official"
+
+        elif model == "gemini-3-pro-image-preview-url":
+            if resolution == "1K":
+                return "gemini-3-pro-image-preview-url"
+            elif resolution == "4K":
+                return "gemini-3-pro-image-preview-4k-url"
+            else:  # 2K（默认）
+                return "gemini-3-pro-image-preview-2k-url"
+
+        # ── 固定端点模型：从 models_config 里取端点，提取模型名段 ──────────
+        from ..models_config import get_model_endpoint
+        endpoint = get_model_endpoint(model)
+        if endpoint:
+            # 端点格式：/v1beta/models/<model-name>:generateContent
+            # 提取 <model-name> 部分
+            match = re.search(r"/models/([^:]+):", endpoint)
+            if match:
+                return match.group(1)
+
+        # ── 兜底：直接用 model ID ──────────────────────────────────────────
+        return model
+
+    # ------------------------------------------------------------------ #
+    #  BaseAPIClient 抽象方法实现                                           #
+    # ------------------------------------------------------------------ #
+
+    def get_endpoint(self, **kwargs) -> str:
+        """固定返回 /v1/chat/completions，模型信息已移入请求体。"""
+        return _ENDPOINT
+
+    def build_request_body(
+        self,
+        prompt: str = "",
+        images: Optional[List[Image.Image]] = None,
+        aspect_ratio: str = "1:1",
+        resolution: str = "2K",
+        model: str = "",
+        **kwargs
+    ) -> Dict[str, Any]:
+        """
+        构建 OpenAI /v1/chat/completions 格式请求体。
+
+        文生图示例输出：
+        {
+            "model": "nano-banana-pro-2k",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": "一个中国女子的OOTD"}
+                    ]
+                }
+            ],
+            "modalities": ["image", "text"],
+            "stream": false,
+            "extra_body": {
+                "google": {
+                    "image_config": {
+                        "aspect_ratio": "16:9",
+                        "image_size": "2K"
+                    }
+                }
+            }
+        }
+
+        图生图时 content 数组追加若干 image_url 块：
+            {
+                "type": "image_url",
+                "image_url": {"url": "data:image/png;base64,<...>"}
+            }
+
+        Args:
+            prompt:       提示词
+            images:       参考图列表（可选，图生图时传入）
+            aspect_ratio: 宽高比，如 "16:9"
+            resolution:   分辨率，如 "2K"
+            model:        已解析好的模型名（由 resolve_model_name 返回）
+        """
+        # ── 构建 content 数组 ─────────────────────────────────────────────
+        content: List[Dict[str, Any]] = []
+
+        # 1. 文本部分（始终在最前）
+        content.append({
+            "type": "text",
+            "text": prompt
+        })
+
+        # 2. 图片部分（图生图时追加，每张图一个 image_url block）
+        if images:
+            for img in images:
+                b64 = encode_image_to_base64(img)
+                content.append({
+                    "type": "image_url",
+                    "image_url": {
+                        "url": f"data:image/png;base64,{b64}"
+                    }
+                })
+
+        # ── 分辨率映射（节点内部值 → API 所需值） ────────────────────────────
+        _resolution_map = {"512": "0.5K", "1K": "1K", "2K": "2K", "4K": "4K"}
+        api_image_size = _resolution_map.get(resolution, resolution)
+
+        # ── 组装完整请求体 ─────────────────────────────────────────────────
+        request_body: Dict[str, Any] = {
+            "model": model,
+            "messages": [
+                {
+                    "role": "user",
+                    "content": content
+                }
+            ],
+            "modalities": ["image", "text"],
+            "stream": False,
+            "extra_body": {
+                "google": {
+                    "image_config": {
+                        "image_size": api_image_size
+                    }
+                }
+            }
+        }
+
+        if aspect_ratio and aspect_ratio != "智能":
+            request_body["extra_body"]["google"]["image_config"]["aspect_ratio"] = aspect_ratio
+
+        return request_body
+
+    def parse_response(self, response: Dict[str, Any]) -> List[Image.Image]:
+        """同步 parse_response，仅为满足抽象基类要求，实际不应被直接调用。"""
+        raise RuntimeError(
+            "parse_response() 不应被直接调用。"
+            "请使用 generate_single_async() 等高级方法。"
+        )
+
+    def get_http_error_message(self, status_code: int, error_message: str) -> Optional[str]:
+        """429 / 503 友好文案。"""
+        if status_code == 429:
+            return (
+                "莫慌!该模型暂时超出速率限制啦\n"
+                "解决方案如下(任意一种):\n"
+                "1.切换当前模型\n"
+                "2.前往后台,修改令牌分组"
+            )
+        if status_code == 503:
+            return (
+                "警报!服务器当前过载!\n"
+                "解决方案如下:\n"
+                "1.摸会儿鱼吧,稍后会恢复,嘿嘿~\n"
+                "2.切换其他模型\n"
+                "3.前往后台,修改令牌分组"
+            )
+        return None
+
+    # ------------------------------------------------------------------ #
+    #  响应解析                                                             #
+    # ------------------------------------------------------------------ #
+
+    async def parse_response_async(
+        self,
+        response: Dict[str, Any],
+        session: Optional[aiohttp.ClientSession] = None
+    ) -> tuple[List[Image.Image], Dict[str, Any]]:
+        """
+        异步解析 /v1/chat/completions 格式响应，提取生成的图像。
+
+        响应结构（OpenAI 格式）：
+        {
+            "choices": [
+                {
+                    "message": {
+                        "role": "assistant",
+                        "content": [
+                            {"type": "text",  "text": "..."},
+                            {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}}
+                            // 或直接 inline_data / inlineData（兼容 Gemini 风格回包）
+                        ]
+                    },
+                    "finish_reason": "stop"
+                }
+            ],
+            "usage": {...}
+        }
+        """
+        format_info: Dict[str, Any] = {
+            "type": None,       # "base64" | "url"
+            "size": 0,
+            "resolution": None,
+            "download_speed": None
+        }
+
+        # ── 错误前置检测 ───────────────────────────────────────────────────
+
+        # 1. usage.completion_tokens == 0 → 风控拦截（对齐 Gemini 的 candidatesTokenCount==0）
+        usage = response.get("usage", {})
+        completion_tokens = usage.get("completion_tokens", -1)
+        if completion_tokens == 0:
+            raise RuntimeError(
+                "Damn!你触发顶级风控啦!还没到生图阶段就被拒了。\n"
+                "赶紧调整一下图片或提示词吧!该情况不会返回图片且正常扣费!下次小心哦~"
+            )
+
+        # 2. finish_reason 不是 "stop" → 安全过滤 / token 超限等
+        choices = response.get("choices", [])
+        if choices:
+            for choice in choices:
+                finish_reason = choice.get("finish_reason", "")
+                if finish_reason and finish_reason != "stop":
+                    raise RuntimeError(
+                        "Ohh no! 生图过程触发风控,图片被拒绝生成!\n"
+                        "可能原因如下:\n"
+                        "1.违禁内容\n"
+                        "2.触发安全过滤器\n"
+                        "3.涉及版权问题\n"
+                        "4. Token超限\n"
+                        "赶紧调整一下图片或提示词吧!该情况不会返回图片且正常扣费!下次小心哦~"
+                    )
+
+        # ── 图像提取 ───────────────────────────────────────────────────────
+        images: List[Image.Image] = []
+        text_responses: List[str] = []
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            for choice in choices:
+                message = choice.get("message", {})
+
+                # ── 优先从 message.images 提取（非标准扩展字段） ──────────────
+                # 部分服务端把图片放在独立的 images 字段，content 同时为 null
+                msg_images = message.get("images") or []
+                for img_part in msg_images:
+                    part_type = img_part.get("type", "")
+                    if part_type == "image_url":
+                        url_obj = img_part.get("image_url", {})
+                        url = url_obj.get("url", "")
+                        if url.startswith("data:"):
+                            try:
+                                _, b64_data = url.split(",", 1)
+                                img = decode_base64_to_pil(b64_data)
+                                images.append(img)
+                                if format_info["type"] is None:
+                                    format_info["type"] = "base64"
+                                    format_info["size"] = len(b64_data) * 3 / 4
+                                    format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                            except Exception:
+                                pass
+                        elif url.startswith("http"):
+                            try:
+                                dl_start = time.time()
+                                async with session.get(url) as img_resp:
+                                    if img_resp.status == 200:
+                                        img_data = await img_resp.read()
+                                        dl_time = time.time() - dl_start
+                                        speed = len(img_data) / dl_time if dl_time > 0 else 0
+                                        img = Image.open(BytesIO(img_data))
+                                        images.append(img)
+                                        if format_info["type"] is None:
+                                            format_info["type"] = "url"
+                                            format_info["size"] = len(img_data)
+                                            format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                                            format_info["download_speed"] = speed
+                            except Exception:
+                                pass
+
+                # ── 再从 message.content 提取（标准 OpenAI 格式） ─────────────
+                # content 为 null 时用空列表兜底，避免 for in None 崩溃
+                raw_content = message.get("content") or []
+
+                # content 可能是字符串（纯文本）或数组（多模态）
+                if isinstance(raw_content, str):
+                    text_responses.append(raw_content)
+                    continue
+
+                for part in raw_content:
+                    part_type = part.get("type", "")
+
+                    # ── 情况 A：OpenAI image_url 格式 ─────────────────────
+                    if part_type == "image_url":
+                        url_obj = part.get("image_url", {})
+                        url = url_obj.get("url", "")
+
+                        if url.startswith("data:"):
+                            # data URI → 直接 base64 解码
+                            # 格式：data:image/png;base64,<data>
+                            try:
+                                header, b64_data = url.split(",", 1)
+                                img = decode_base64_to_pil(b64_data)
+                                images.append(img)
+                                if format_info["type"] is None:
+                                    format_info["type"] = "base64"
+                                    format_info["size"] = len(b64_data) * 3 / 4
+                                    format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                            except Exception:
+                                pass
+
+                        elif url.startswith("http"):
+                            # 远程 URL → 异步下载
+                            try:
+                                dl_start = time.time()
+                                async with session.get(url) as img_resp:
+                                    if img_resp.status == 200:
+                                        img_data = await img_resp.read()
+                                        dl_time = time.time() - dl_start
+                                        speed = len(img_data) / dl_time if dl_time > 0 else 0
+                                        img = Image.open(BytesIO(img_data))
+                                        images.append(img)
+                                        if format_info["type"] is None:
+                                            format_info["type"] = "url"
+                                            format_info["size"] = len(img_data)
+                                            format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                                            format_info["download_speed"] = speed
+                            except Exception:
+                                pass
+
+                    # ── 情况 B：Gemini 风格 inline_data / inlineData（兼容） ─
+                    elif part_type in ("inline_data", "inlineData") or \
+                            "inline_data" in part or "inlineData" in part:
+                        inline_key = "inline_data" if "inline_data" in part else "inlineData"
+                        inline = part.get(inline_key, {})
+                        b64_data = inline.get("data", "")
+                        if b64_data:
+                            try:
+                                img = decode_base64_to_pil(b64_data)
+                                images.append(img)
+                                if format_info["type"] is None:
+                                    format_info["type"] = "base64"
+                                    format_info["size"] = len(b64_data) * 3 / 4
+                                    format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                            except Exception:
+                                pass
+
+                    # ── 情况 C：text 中嵌套 URL（markdown 或纯链接） ─────────
+                    elif part_type == "text":
+                        text = part.get("text", "")
+                        text_responses.append(text)
+
+                        # markdown 图片链接：![alt](url)
+                        urls = re.findall(r'!\[.*?\]\((https?://[^\)]+)\)', text)
+                        if not urls:
+                            urls = re.findall(r'https?://[^\s<>"{}|\\^`\[\]]+', text)
+
+                        for url in urls:
+                            try:
+                                dl_start = time.time()
+                                async with session.get(url) as img_resp:
+                                    if img_resp.status == 200:
+                                        img_data = await img_resp.read()
+                                        dl_time = time.time() - dl_start
+                                        speed = len(img_data) / dl_time if dl_time > 0 else 0
+                                        img = Image.open(BytesIO(img_data))
+                                        images.append(img)
+                                        if format_info["type"] is None:
+                                            format_info["type"] = "url"
+                                            format_info["size"] = len(img_data)
+                                            format_info["resolution"] = f"{img.size[0]}x{img.size[1]}"
+                                            format_info["download_speed"] = speed
+                            except Exception:
+                                pass
+
+        except RuntimeError:
+            raise
+        except Exception as e:
+            raise RuntimeError(f"解析 API 响应失败: {str(e)}")
+        finally:
+            if close_session:
+                await session.close()
+
+        # ── 3. 无图像但有文本 → API 拒绝说明 ─────────────────────────────
+        if not images and text_responses:
+            combined = "\n".join(text_responses)
+            raise RuntimeError(
+                f"API 拒绝响应\n\n"
+                f"API 返回说明：\n{combined}\n\n"
+                f"建议：\n"
+                f"  - 根据上述说明调整请求内容\n"
+                f"  - 确保提示词和参考图符合使用规范"
+            )
+
+        if not images:
+            raise RuntimeError("API 响应中未找到生成的图像")
+
+        return images, format_info
+
+    # ------------------------------------------------------------------ #
+    #  核心生成方法（接口与 GeminiAPIClient 保持一致，节点可无缝切换）        #
+    # ------------------------------------------------------------------ #
+
+    async def generate_single_async(
+        self,
+        prompt: str,
+        model: str,
+        resolution: str,
+        aspect_ratio: str,
+        images: Optional[List[Image.Image]] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+        task_index: Optional[int] = None,
+        total_tasks: Optional[int] = None,
+        debug: bool = False,
+        debug_request: bool = False,
+        enable_grounding: bool = False,      # 保留签名兼容，OpenAI 格式暂不使用
+        enable_image_search: bool = False    # 保留签名兼容，OpenAI 格式暂不使用
+    ) -> tuple[List[Image.Image], Dict[str, Any]]:
+        """
+        单次异步生成请求（OpenAI /v1/chat/completions 格式）。
+
+        Args:
+            prompt:        提示词
+            model:         节点选中的模型 ID（将自动解析为实际模型名）
+            resolution:    分辨率
+            aspect_ratio:  宽高比
+            images:        参考图列表（图生图时传入）
+            session:       复用的 aiohttp 会话
+            task_index:    任务序号（批量时用于日志）
+            total_tasks:   总任务数（批量时用于日志）
+            debug:         打印完整 API 响应
+            debug_request: 打印请求体（base64 自动截断）
+
+        Returns:
+            (生成的图像列表, 计时信息字典)
+        """
+        import json
+
+        total_start = time.time()
+        task_prefix = f"[{task_index}/{total_tasks}]" if task_index is not None and total_tasks else ""
+
+        # ── 1. 解析模型名 & 构建请求体 ────────────────────────────────────
+        build_start = time.time()
+        resolved_model = self.resolve_model_name(model, resolution)
+        endpoint = self.get_endpoint()
+
+        request_body = self.build_request_body(
+            prompt=prompt,
+            images=images,
+            aspect_ratio=aspect_ratio,
+            resolution=resolution,
+            model=resolved_model
+        )
+        build_time = time.time() - build_start
+
+        # ── 调试：打印请求体 ───────────────────────────────────────────────
+        if debug_request:
+            import json as _json
+            def _shorten_b64(obj):
+                if isinstance(obj, dict):
+                    return {k: _shorten_b64(v) for k, v in obj.items()}
+                if isinstance(obj, list):
+                    return [_shorten_b64(i) for i in obj]
+                if isinstance(obj, str):
+                    if obj.startswith("data:"):
+                        header, _, data = obj.partition(",")
+                        return f"{header},<base64 {len(data)} chars>"
+                    if len(obj) > 200 and all(
+                        c in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/="
+                        for c in obj[:64]
+                    ):
+                        return f"<base64 {len(obj)} chars>"
+                return obj
+            print(
+                f"\n{'='*60}\n"
+                f"[请求体日志] 任务 {task_prefix or '?'}\n"
+                f"端点: {self.base_url}{endpoint}\n"
+                f"{_json.dumps(_shorten_b64(request_body), ensure_ascii=False, indent=2)}\n"
+                f"{'='*60}\n"
+            )
+
+        # ── 2. 计算请求体大小 ─────────────────────────────────────────────
+        request_size = len(json.dumps(request_body).encode("utf-8"))
+        size_str = (
+            f"{request_size / 1024:.2f}KB"
+            if request_size < 1024 * 1024
+            else f"{request_size / (1024 * 1024):.2f}MB"
+        )
+
+        # ── 3. 发送请求（Bearer Token 认证） ─────────────────────────────
+        request_start = time.time()
+        try:
+            response = await self.request_async(
+                endpoint,
+                request_body,
+                session,
+                use_bearer_token=True
+            )
+        except Exception as e:
+            request_time = time.time() - request_start
+            error_first_line = str(e).split("\n")[0]
+            print(f"{task_prefix} 请求 {size_str} → API {request_time:.1f}s → 失败: {error_first_line} ✗")
+            raise
+
+        request_time = time.time() - request_start
+
+        # ── 调试：打印完整响应 ─────────────────────────────────────────────
+        if debug:
+            import json as _json
+            def _shorten_b64(obj):
+                if isinstance(obj, dict):
+                    return {k: _shorten_b64(v) for k, v in obj.items()}
+                if isinstance(obj, list):
+                    return [_shorten_b64(i) for i in obj]
+                if isinstance(obj, str):
+                    if obj.startswith("data:"):
+                        header, _, data = obj.partition(",")
+                        return f"{header},<base64 {len(data)} chars>"
+                    if len(obj) > 200 and all(
+                        c in "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/="
+                        for c in obj[:64]
+                    ):
+                        return f"<base64 {len(obj)} chars>"
+                return obj
+            print(
+                f"\n{'='*60}\n"
+                f"[调试日志] 任务 {task_prefix or '?'} 完整 API 响应：\n"
+                f"{_json.dumps(_shorten_b64(response), ensure_ascii=False, indent=2)}\n"
+                f"{'='*60}\n"
+            )
+
+        # ── 4. 解析响应 ───────────────────────────────────────────────────
+        parse_start = time.time()
+        try:
+            result_images, format_info = await self.parse_response_async(response, session)
+        except Exception as e:
+            parse_time = time.time() - parse_start
+            error_first_line = str(e).split("\n")[0]
+            print(f"{task_prefix} 请求 {size_str} → API {request_time:.1f}s → 解析失败: {error_first_line} ✗")
+            raise
+
+        parse_time = time.time() - parse_start
+
+        # ── 5. 单行日志输出 ───────────────────────────────────────────────
+        img_size = format_info.get("size", 0)
+        img_size_str = (
+            f"{img_size / 1024:.2f}KB"
+            if img_size < 1024 * 1024
+            else f"{img_size / (1024 * 1024):.2f}MB"
+        )
+
+        if format_info.get("type") == "base64":
+            download_info = f"Base64 {img_size_str} ({parse_time:.1f}s)"
+        elif format_info.get("type") == "url":
+            speed = format_info.get("download_speed", 0)
+            download_info = f"URL {img_size_str} ({parse_time:.1f}s, {speed / (1024*1024):.1f}MB/s)"
+        else:
+            download_info = img_size_str
+
+        timing = response.get("_timing", {})
+        net_connect = timing.get("connect_time")
+        net_download = timing.get("download_time")
+        if net_connect is not None and net_download is not None:
+            net_str = f" | 连接 {net_connect:.2f}s | 下载 {net_download:.2f}s"
+        else:
+            net_str = ""
+        print(f"{task_prefix} 请求 {size_str} → API {request_time:.1f}s → {download_info} ✓{net_str}")
+
+        total_time = time.time() - total_start
+        timing_info = {
+            "build_time": build_time,
+            "request_time": request_time,
+            "parse_time": parse_time,
+            "total_time": total_time,
+            "format_type": format_info.get("type", "unknown")
+        }
+
+        return result_images, timing_info
+
+    # ------------------------------------------------------------------ #
+    #  批量 & 同步接口（与 GeminiAPIClient 接口签名一致）                    #
+    # ------------------------------------------------------------------ #
+
+    async def generate_batch_async(
+        self,
+        prompt: str,
+        model: str,
+        resolution: str,
+        aspect_ratio: str,
+        batch_size: int,
+        images: Optional[List[Image.Image]] = None,
+        progress_callback: Optional[Callable[[int, int, bool, Optional[str]], None]] = None,
+        debug: bool = False,
+        debug_request: bool = False,
+        enable_grounding: bool = False,
+        enable_image_search: bool = False
+    ) -> List[Image.Image]:
+        """批量全并发生成（单提示词 × batch_size 张）。"""
+        import asyncio
+
+        all_images: List[Image.Image] = []
+        completed = 0
+        success_count = 0
+        fail_count = 0
+        first_error = None
+
+        max_concurrent = 10
+        num_batches = (batch_size + max_concurrent - 1) // max_concurrent
+
+        print(f"OpenAIClient: 批量生成 {batch_size} 张，并发数: {max_concurrent}，分 {num_batches} 批")
+
+        connector = aiohttp.TCPConnector(ssl=False, limit=0, limit_per_host=0)
+
+        async with aiohttp.ClientSession(connector=connector) as session:
+            for batch_idx in range(num_batches):
+                batch_start = batch_idx * max_concurrent
+                batch_end = min(batch_start + max_concurrent, batch_size)
+                batch_count = batch_end - batch_start
+
+                if num_batches > 1:
+                    print(f"OpenAIClient: 第 {batch_idx + 1}/{num_batches} 批 ({batch_start + 1}-{batch_end})")
+
+                tasks = [
+                    asyncio.create_task(
+                        self.generate_single_async(
+                            prompt=prompt,
+                            model=model,
+                            resolution=resolution,
+                            aspect_ratio=aspect_ratio,
+                            images=images,
+                            session=session,
+                            task_index=batch_start + i + 1,
+                            total_tasks=batch_size,
+                            debug=debug,
+                            debug_request=debug_request
+                        ),
+                        name=f"task_{batch_start + i}"
+                    )
+                    for i in range(batch_count)
+                ]
+
+                batch_images: List[Image.Image] = []
+
+                for coro in asyncio.as_completed(tasks):
+                    completed += 1
+                    try:
+                        result_imgs, _ = await coro
+                        for img in result_imgs:
+                            batch_images.append(img)
+                            all_images.append(img)
+                        success_count += 1
+                        if progress_callback:
+                            progress_callback(completed, batch_size, True, None)
+                        print(f"OpenAIClient: 任务 {completed}/{batch_size} 成功 ✓")
+                    except Exception as e:
+                        fail_count += 1
+                        if first_error is None:
+                            first_error = e
+                        if progress_callback:
+                            progress_callback(completed, batch_size, False, str(e))
+                        print(f"OpenAIClient: 任务 {completed}/{batch_size} 失败 ✗")
+
+                if batch_images:
+                    print(f"OpenAIClient: 第 {batch_idx + 1} 批完成，生成 {len(batch_images)} 张")
+                    import gc
+                    gc.collect()
+                    await asyncio.sleep(0.1)
+
+                batch_images = []
+
+        if not all_images:
+            if first_error:
+                raise first_error
+            raise RuntimeError(f"批量生成失败，{fail_count} 个请求全部失败")
+
+        print(f"OpenAIClient: 批量完成，成功 {success_count}/{batch_size}，失败 {fail_count}")
+        return all_images
+
+    def generate_sync(
+        self,
+        prompt: str,
+        model: str,
+        resolution: str,
+        aspect_ratio: str,
+        batch_size: int,
+        images: Optional[List[Image.Image]] = None,
+        progress_callback: Optional[Callable[[int, int, bool, Optional[str]], None]] = None,
+        debug: bool = False,
+        debug_request: bool = False,
+        enable_grounding: bool = False,
+        enable_image_search: bool = False
+    ) -> List[Image.Image]:
+        """同步生成接口（用于 ComfyUI 节点，接口与 GeminiAPIClient 完全一致）。"""
+        coro = self.generate_batch_async(
+            prompt=prompt,
+            model=model,
+            resolution=resolution,
+            aspect_ratio=aspect_ratio,
+            batch_size=batch_size,
+            images=images,
+            progress_callback=progress_callback,
+            debug=debug,
+            debug_request=debug_request,
+            enable_grounding=enable_grounding,
+            enable_image_search=enable_image_search
+        )
+        return self.run_async_in_thread(coro)
@@ -0,0 +1,195 @@
+"""
+Seedance 视频生成客户端
+使用 new-api 原生格式：POST /v1/video/generations → GET /v1/video/generations/{task_id}
+"""
+
+import asyncio
+import json
+import os
+from typing import Any, Callable, Dict, Optional
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+
+class SeedanceClient:
+    """Seedance 视频生成客户端（new-api 原生三段式）"""
+
+    # 提交任务
+    CREATE_ENDPOINT = "/v1/video/generations"
+    # 查询任务状态：{task_id} 占位
+    STATUS_ENDPOINT = "/v1/video/generations/{task_id}"
+
+    POLL_INITIAL_INTERVAL = 4   # 首次轮询等待秒数
+    POLL_MAX_INTERVAL = 15      # 最大轮询间隔秒数
+
+    # new-api 返回的成功状态值
+    SUCCESS_STATUSES = {"succeeded", "success", "completed", "done", "finished"}
+    FAILURE_STATUSES = {"failed", "fail", "error", "expired"}
+
+    def __init__(self):
+        self.api_key = get_api_key_or_raise()
+        self.base_url = "https://api.o1key.com"
+
+    def _headers(self) -> Dict[str, str]:
+        return {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json",
+        }
+
+    # ── 1. 提交任务 ────────────────────────────────────────────────────
+
+    async def submit_async(
+        self,
+        body: Dict[str, Any],
+        session: aiohttp.ClientSession,
+    ) -> str:
+        """提交视频生成任务，返回 task_id"""
+        url = f"{self.base_url}{self.CREATE_ENDPOINT}"
+        check_interrupt()
+        resp = await run_with_interrupt(async_request_with_retry(
+            session, "POST", url, json=body, headers=self._headers(), prefix="Seedance 提交: "
+        ))
+        check_interrupt()
+        text = await resp.text()
+        data = json.loads(text)
+
+        # new-api 返回字段：id / task_id
+        task_id = data.get("id") or data.get("task_id")
+        if not task_id:
+            raise RuntimeError(f"API 未返回任务 ID，响应：{data}")
+        return task_id
+
+    # ── 2. 轮询状态 ────────────────────────────────────────────────────
+
+    async def poll_async(
+        self,
+        task_id: str,
+        session: aiohttp.ClientSession,
+        on_progress: Optional[Callable[[int], None]] = None,
+    ) -> str:
+        """轮询任务状态，成功后返回视频 URL"""
+        url = f"{self.base_url}{self.STATUS_ENDPOINT.format(task_id=task_id)}"
+        interval = self.POLL_INITIAL_INTERVAL
+
+        while True:
+            check_interrupt()
+            async with session.get(url, headers=self._headers()) as resp:
+                text = await resp.text()
+                if resp.status != 200:
+                    try:
+                        err = json.loads(text)
+                        msg = (err.get("error", {}).get("message")
+                               or err.get("message")
+                               or text)
+                    except Exception:
+                        msg = text
+                    raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                result = json.loads(text)
+
+            # new-api 包装格式：真实数据在 result["data"] 里
+            inner = result.get("data") or result
+
+            status = extract_status(result)
+
+            # 解析进度
+            progress_pct = extract_progress(result)
+
+            print(f"[Seedance] 生成中 {progress_pct}%")
+            if on_progress:
+                on_progress(progress_pct)
+
+            if is_success_status(status):
+                # 响应结构：result["data"] = inner，inner["data"] = platform_data
+                # 视频 URL 在 inner["result_url"] 或 inner["data"]["content"]["video_url"]
+                platform_data = inner.get("data") or {}
+                content = platform_data.get("content") or {}
+                video_url = (
+                    inner.get("result_url")
+                    or content.get("video_url")
+                    or platform_data.get("video_url")
+                    or inner.get("url")
+                )
+                if not video_url:
+                    raise RuntimeError(f"任务成功但未找到视频 URL，响应：{result}")
+                # 末帧图片 URL 在 inner["data"]["content"]["last_frame_url"]
+                last_frame_url = (
+                    content.get("last_frame_url")
+                    or platform_data.get("last_frame_url")
+                    or inner.get("last_frame_url")
+                )
+                return video_url, last_frame_url
+
+            if is_failure_status(status, result):
+                reason = extract_error_message(result)
+                raise RuntimeError(f"视频生成失败：{reason}")
+
+            await interruptible_sleep(interval)
+            interval = min(interval * 1.5, self.POLL_MAX_INTERVAL)
+
+    # ── 3. 下载视频 ────────────────────────────────────────────────────
+
+    async def download_async(
+        self,
+        video_url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+    ) -> str:
+        """下载视频到本地，返回本地路径"""
+        print(f"[Seedance] 下载视频...")
+        check_interrupt()
+        async with session.get(video_url, allow_redirects=True) as resp:
+            if resp.status != 200:
+                raise RuntimeError(f"视频下载失败 ({resp.status})")
+            os.makedirs(os.path.dirname(save_path), exist_ok=True)
+            with open(save_path, "wb") as f:
+                async for chunk in resp.content.iter_chunked(8192):
+                    check_interrupt()
+                    f.write(chunk)
+        return save_path
+
+    # ── 全流程入口（供节点调用）────────────────────────────────────────
+
+    async def generate_async(
+        self,
+        body: Dict[str, Any],
+        save_path: str,
+        on_stage: Optional[Callable[[str], None]] = None,
+        on_progress: Optional[Callable[[int], None]] = None,
+    ) -> tuple:
+        """提交 → 轮询 → 下载，返回 (本地视频路径, 末帧图片URL或None)"""
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 提交
+            check_interrupt()
+            if on_stage:
+                on_stage("submitting")
+            task_id = await self.submit_async(body, session)
+            print(f"[Seedance] 任务已提交 → {task_id}")
+            if on_stage:
+                on_stage(f"submitted:{task_id}")
+
+            # 轮询
+            video_url, last_frame_url = await self.poll_async(task_id, session, on_progress=on_progress)
+
+            # 下载
+            if on_stage:
+                on_stage("downloading")
+            path = await self.download_async(video_url, save_path, session)
+
+            if on_stage:
+                on_stage("done")
+            return path, last_frame_url
@@ -0,0 +1,530 @@
+"""
+Sora 视频生成 API 客户端
+提供视频创建、状态轮询、视频下载功能
+"""
+
+import asyncio
+import base64
+import json
+import os
+import time
+from typing import Any, Callable, Dict, List, Optional
+
+import aiohttp
+
+from .base_client import BaseAPIClient
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..utils.image_utils import encode_image_to_base64
+
+
+def _translate_error_message(msg: str) -> str:
+    """将 API 返回的已知英文错误信息翻译为中文友好提示"""
+    if "people-in-user-uploads" in msg or (
+        "moderation" in msg and "inputs" in msg
+    ):
+        return "上传的参考图片中包含了真实人物【官方风控】，请尝试使用其他办法绕开。"
+    return msg
+
+
+class SoraClient(BaseAPIClient):
+    """
+    Sora 视频生成客户端
+
+    工作流程：
+    1. create_video  → POST /v1/videos  (提交生成任务)
+    2. poll_status   → GET  /v1/videos/{id}  (轮询直到完成/失败)
+    3. download_video→ GET  /v1/videos/{id}/content (下载视频文件)
+    """
+
+    CREATE_ENDPOINT = "/v1/videos"
+    STATUS_ENDPOINT = "/v1/videos/{video_id}"
+    CONTENT_ENDPOINT = "/v1/videos/{video_id}/content"
+
+    POLL_INITIAL_INTERVAL = 3
+    POLL_MAX_INTERVAL = 15
+
+    def __init__(self):
+        api_key = get_api_key_or_raise()
+        base_url = get_api_base_url()
+        super().__init__(base_url=base_url, api_key=api_key)
+
+    # ------------------------------------------------------------------
+    # BaseAPIClient 抽象方法实现（本客户端主要使用自定义方法）
+    # ------------------------------------------------------------------
+
+    def get_endpoint(self, **kwargs) -> str:
+        return self.CREATE_ENDPOINT
+
+    def build_request_body(self, **kwargs) -> Dict[str, Any]:
+        return {}
+
+    def parse_response(self, response: Dict[str, Any]) -> Any:
+        return response
+
+    # ------------------------------------------------------------------
+    # 核心异步方法
+    # ------------------------------------------------------------------
+
+    async def create_video_async(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int = 4,
+        size: str = "720x1280",
+        input_reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        """
+        提交视频生成任务
+
+        格式策略（根据抓包确认）：
+        - 无参考图片：application/json
+        - 有参考图片：multipart/form-data，input_reference 以 PNG 文件上传
+
+        注意：seed 不被上游 API 接受，仅在 ComfyUI 节点侧用于缓存刷新
+
+        Returns:
+            API 响应 JSON，包含 video id 和初始状态
+        """
+        url = f"{self.base_url}{self.CREATE_ENDPOINT}"
+        headers = {"Authorization": f"Bearer {self.api_key}"}
+
+        # ============================================================
+        # ⚠️  已验证可用的标准请求方案，请勿随意修改！（2026-02-28）
+        # ============================================================
+        # 经多轮调试确认：
+        #   - 有图片：必须使用 multipart/form-data，input_reference 以 PNG 文件上传
+        #     · filename="reference.png", content_type="image/png"（与抓包一致）
+        #     · 不可改为 application/json + base64 → 400 "expected a file, got a string"
+        #     · 不可改为 application/json + data URI → 500 upstream error
+        #     · 不可改为 multipart + image/jpeg     → 400 "Inpaint image must match..."（尺寸校验失败）
+        #   - 无图片：使用 application/json，已验证成功
+        # ============================================================
+        if input_reference_bytes:
+            if len(input_reference_bytes) > self.max_request_size:
+                raise ValueError(
+                    f"参考图片约 {len(input_reference_bytes) / 1024 / 1024:.1f}MB，"
+                    f"超过 {self.max_request_size / 1024 / 1024:.0f}MB 限制，请使用较小的图片"
+                )
+            # ⚠️ 有图片：multipart/form-data + PNG 文件上传（唯一验证成功的方案）
+            form = aiohttp.FormData()
+            form.add_field("prompt", prompt)
+            form.add_field("model", model)
+            form.add_field("seconds", str(seconds))
+            form.add_field("size", size)
+            form.add_field(
+                "input_reference",
+                input_reference_bytes,
+                filename="reference.png",   # ⚠️ 不可改文件名/扩展名
+                content_type="image/png",   # ⚠️ 不可改为 image/jpeg
+            )
+            send_kwargs: Dict[str, Any] = {"data": form, "headers": headers}
+        else:
+            # ⚠️ 无图片：application/json（已验证成功）
+            body: Dict[str, Any] = {
+                "model": model,
+                "prompt": prompt,
+                "seconds": str(seconds),
+                "size": size,
+            }
+            send_kwargs = {"json": body, "headers": headers}
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            async with session.post(url, **send_kwargs) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    error_message = self._extract_error_message(error_text, response.status)
+                    raise RuntimeError(error_message)
+
+                resp_json = await response.json()
+                return resp_json
+
+        finally:
+            if close_session:
+                await session.close()
+
+    async def poll_video_status_async(
+        self,
+        video_id: str,
+        progress_callback: Optional[Callable[[int, float], None]] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        """
+        轮询视频生成状态，直到完成或失败
+
+        Args:
+            video_id: 视频任务 ID
+            progress_callback: 进度回调 (progress_percent, elapsed_seconds)
+            session: aiohttp 会话
+
+        Returns:
+            最终状态的 API 响应
+
+        Raises:
+            RuntimeError: 生成失败
+        """
+        url = f"{self.base_url}{self.STATUS_ENDPOINT.format(video_id=video_id)}"
+        headers = self.get_headers(use_bearer_token=True)
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        interval = self.POLL_INITIAL_INTERVAL
+
+        try:
+            while True:
+                async with session.get(url, headers=headers) as response:
+                    if response.status != 200:
+                        error_text = await response.text()
+                        error_message = self._extract_error_message(error_text, response.status)
+                        raise RuntimeError(error_message)
+
+                    data = await response.json()
+
+                # status 兼容大小写：queued / in_progress / IN_PROGRESS / completed / COMPLETED
+                status = data.get("status", "").lower()
+
+                # progress 兼容整数 (30) 和字符串 ("30%") 两种格式
+                progress_raw = data.get("progress", 0)
+                if isinstance(progress_raw, str):
+                    try:
+                        progress = int(progress_raw.rstrip("%").strip())
+                    except ValueError:
+                        progress = 0
+                else:
+                    progress = int(progress_raw) if progress_raw else 0
+
+                if progress_callback:
+                    progress_callback(progress)
+
+                if status == "completed":
+                    return data
+
+                if status == "failed":
+                    error_info = data.get("error", {})
+                    error_msg = error_info.get("message", "未知错误") if isinstance(error_info, dict) else str(error_info)
+                    error_msg = _translate_error_message(error_msg)
+                    raise RuntimeError(f"视频生成失败: {error_msg}")
+
+                await asyncio.sleep(interval)
+                interval = min(interval * 1.5, self.POLL_MAX_INTERVAL)
+
+        finally:
+            if close_session:
+                await session.close()
+
+    async def download_video_async(
+        self,
+        video_id: str,
+        save_path: str,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> str:
+        """
+        下载生成的视频文件
+
+        处理两种情况：
+        1. 响应为重定向或 JSON 含下载 URL → 跟随下载
+        2. 响应为二进制视频流 → 直接保存
+
+        Returns:
+            保存的文件路径
+        """
+        url = f"{self.base_url}{self.CONTENT_ENDPOINT.format(video_id=video_id)}"
+        headers = self.get_headers(use_bearer_token=True)
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            async with session.get(url, headers=headers, allow_redirects=True) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    error_message = self._extract_error_message(error_text, response.status)
+                    raise RuntimeError(f"视频下载失败: {error_message}")
+
+                content_type = response.headers.get("Content-Type", "")
+
+                if "application/json" in content_type:
+                    data = await response.json()
+                    download_url = data.get("url") or data.get("download_url")
+                    if not download_url:
+                        raise RuntimeError("视频下载失败: 响应中未找到下载链接")
+                    await self._download_from_url(download_url, save_path, session)
+                else:
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in response.content.iter_chunked(8192):
+                            f.write(chunk)
+
+            return save_path
+
+        finally:
+            if close_session:
+                await session.close()
+
+    # ------------------------------------------------------------------
+    # 同步包装
+    # ------------------------------------------------------------------
+
+    def generate_video_sync(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_path: str,
+        input_reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        progress_callback: Optional[Callable[[int, float], None]] = None,
+        on_stage: Optional[Callable[[str], None]] = None,
+    ) -> str:
+        """
+        同步执行完整的视频生成流程（创建 → 轮询 → 下载）
+
+        Args:
+            on_stage: 阶段回调，用于打印状态切换信息
+
+        Returns:
+            保存的视频文件路径
+        """
+
+        async def _run():
+            connector = aiohttp.TCPConnector(ssl=False, limit=0)
+            async with aiohttp.ClientSession(connector=connector) as session:
+                # 1. 提交任务
+                if on_stage:
+                    on_stage("submitting")
+                result = await self.create_video_async(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=size,
+                    input_reference_bytes=input_reference_bytes,
+                    seed=seed,
+                    session=session,
+                )
+                video_id = result.get("id")
+                if not video_id:
+                    raise RuntimeError("API 未返回视频任务 ID")
+
+                if on_stage:
+                    on_stage(f"submitted:{video_id}")
+
+                # 2. 轮询状态
+                if on_stage:
+                    on_stage("polling")
+                await self.poll_video_status_async(
+                    video_id=video_id,
+                    progress_callback=progress_callback,
+                    session=session,
+                )
+
+                # 3. 下载视频
+                if on_stage:
+                    on_stage("downloading")
+                path = await self.download_video_async(
+                    video_id=video_id,
+                    save_path=save_path,
+                    session=session,
+                )
+
+                if on_stage:
+                    on_stage("done")
+                return path
+
+        return self.run_async_in_thread(_run())
+
+    async def _generate_one_video_async(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_path: str,
+        input_reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> str:
+        """
+        异步生成单个视频（创建 → 轮询 → 下载）
+
+        Returns:
+            保存的视频文件路径
+        """
+        result = await self.create_video_async(
+            prompt=prompt,
+            model=model,
+            seconds=seconds,
+            size=size,
+            input_reference_bytes=input_reference_bytes,
+            seed=seed,
+            session=session,
+        )
+        video_id = result.get("id")
+        if not video_id:
+            raise RuntimeError("API 未返回视频任务 ID")
+
+        await self.poll_video_status_async(video_id=video_id, session=session)
+        path = await self.download_video_async(
+            video_id=video_id, save_path=save_path, session=session
+        )
+        return path
+
+    async def generate_batch_videos_async(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_paths: List[str],
+        input_reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        progress_callback: Optional[Callable[[int, int, bool, Optional[str]], None]] = None,
+    ) -> List[str]:
+        """
+        并发生成多个视频
+
+        Args:
+            prompt: 提示词
+            model: 模型名称
+            seconds: 视频时长（秒）
+            size: 分辨率
+            save_paths: 各视频的保存路径列表，长度决定并发数量
+            input_reference_bytes: 参考图片字节（可选）
+            seed: 随机种子（仅节点侧使用）
+            progress_callback: 进度回调 (current, total, success, error_msg)
+
+        Returns:
+            成功生成的视频路径列表
+        """
+        batch_size = len(save_paths)
+        connector = aiohttp.TCPConnector(ssl=False, limit=0)
+
+        async with aiohttp.ClientSession(connector=connector) as session:
+            tasks = [
+                self._generate_one_video_async(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=size,
+                    save_path=save_paths[i],
+                    input_reference_bytes=input_reference_bytes,
+                    seed=seed,
+                    session=session,
+                )
+                for i in range(batch_size)
+            ]
+            results = await asyncio.gather(*tasks, return_exceptions=True)
+
+        completed = 0
+        paths: List[str] = []
+        first_error = None
+        for i, result in enumerate(results):
+            if isinstance(result, Exception):
+                error_msg = str(result)
+                print(f"Sora: 第 {i + 1} 个视频生成失败")
+                print(f"原始错误详情:\n{error_msg}")
+                if first_error is None:
+                    first_error = result
+                if progress_callback:
+                    progress_callback(i + 1, batch_size, False, error_msg)
+            else:
+                completed += 1
+                paths.append(result)
+                if progress_callback:
+                    progress_callback(completed, batch_size, True, None)
+
+        if not paths:
+            if first_error:
+                raise first_error
+            raise RuntimeError(f"批量视频生成失败，{batch_size} 个任务全部失败")
+
+        return paths
+
+    def generate_batch_videos_sync(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_paths: List[str],
+        input_reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        progress_callback: Optional[Callable[[int, int, bool, Optional[str]], None]] = None,
+    ) -> List[str]:
+        """
+        同步并发生成多个视频（用于 ComfyUI 节点）
+
+        Args:
+            save_paths: 各视频的保存路径列表，长度决定并发数量
+
+        Returns:
+            成功生成的视频路径列表
+        """
+        coro = self.generate_batch_videos_async(
+            prompt=prompt,
+            model=model,
+            seconds=seconds,
+            size=size,
+            save_paths=save_paths,
+            input_reference_bytes=input_reference_bytes,
+            seed=seed,
+            progress_callback=progress_callback,
+        )
+        return self.run_async_in_thread(coro)
+
+    # ------------------------------------------------------------------
+    # 内部辅助方法
+    # ------------------------------------------------------------------
+
+    async def _download_from_url(
+        self,
+        url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+    ) -> None:
+        """从给定 URL 下载文件到本地路径"""
+        os.makedirs(os.path.dirname(save_path), exist_ok=True)
+        async with session.get(url) as response:
+            if response.status != 200:
+                raise RuntimeError(f"从下载链接获取视频失败 (状态码: {response.status})")
+            with open(save_path, "wb") as f:
+                async for chunk in response.content.iter_chunked(8192):
+                    f.write(chunk)
+
+    @staticmethod
+    def _extract_error_message(error_text: str, status_code: int) -> str:
+        """从错误响应中提取可读的错误信息"""
+        error_message = error_text
+        try:
+            error_json = json.loads(error_text)
+            if "error" in error_json:
+                if isinstance(error_json["error"], dict):
+                    error_message = error_json["error"].get("message", error_text)
+                else:
+                    error_message = str(error_json["error"])
+            elif "message" in error_json:
+                error_message = error_json["message"]
+        except (json.JSONDecodeError, KeyError):
+            pass
+
+        status_hints = {
+            400: "请求参数错误 (400)",
+            401: "认证失败 (401)，请检查 API 密钥",
+            403: "权限不足 (403)，请检查账户权限或余额",
+            429: "请求频率超限 (429)，请稍后重试",
+            503: "服务暂时不可用 (503)，请稍后重试",
+            504: "请求超时 (504)，请稍后重试",
+        }
+        hint = status_hints.get(status_code, f"API 请求失败 (状态码: {status_code})")
+        return f"{hint}\nAPI 返回: {error_message}"
@@ -0,0 +1,510 @@
+"""
+Veo 视频生成 API 客户端
+提供视频创建、状态轮询、视频下载功能
+"""
+
+import asyncio
+import base64
+import json
+import os
+import time
+from typing import Any, Callable, Dict, List, Optional
+
+import aiohttp
+
+from .base_client import BaseAPIClient
+from ..utils.config import get_api_key_or_raise, get_api_base_url
+from ..utils.image_utils import encode_image_to_base64
+
+
+class VeoClient(BaseAPIClient):
+    """
+    Veo 视频生成客户端
+
+    工作流程：
+    1. create_video  → POST /v1/videos  (提交生成任务)
+    2. poll_status   → GET  /v1/videos/{id}  (轮询直到完成/失败)
+    3. download_video→ GET  /v1/videos/{id}/content (下载视频文件)
+    """
+
+    CREATE_ENDPOINT = "/v1/videos"
+    STATUS_ENDPOINT = "/v1/videos/{video_id}"
+    CONTENT_ENDPOINT = "/v1/videos/{video_id}/content"
+
+    POLL_INITIAL_INTERVAL = 3
+    POLL_MAX_INTERVAL = 15
+
+    def __init__(self):
+        api_key = get_api_key_or_raise()
+        base_url = get_api_base_url()
+        super().__init__(base_url=base_url, api_key=api_key)
+
+    # ------------------------------------------------------------------
+    # BaseAPIClient 抽象方法实现
+    # ------------------------------------------------------------------
+
+    def get_endpoint(self, **kwargs) -> str:
+        return self.CREATE_ENDPOINT
+
+    def build_request_body(self, **kwargs) -> Dict[str, Any]:
+        return {}
+
+    def parse_response(self, response: Dict[str, Any]) -> Any:
+        return response
+
+    # ------------------------------------------------------------------
+    # 核心异步方法
+    # ------------------------------------------------------------------
+
+    async def create_video_async(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int = 8,
+        size: str = "720x1280",
+        first_frame_bytes: Optional[bytes] = None,
+        last_frame_bytes: Optional[bytes] = None,
+        reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        """
+        提交视频生成任务
+
+        格式策略：
+        - 无参考图片：application/json
+        - 有参考图片：multipart/form-data，图片以 PNG 文件上传
+
+        Args:
+            prompt: 提示词
+            model: 模型名称
+            seconds: 视频时长（秒）
+            size: 分辨率
+            first_frame_bytes: 首帧图片字节
+            last_frame_bytes: 尾帧图片字节
+            reference_bytes: 参考图片字节
+            seed: 随机种子
+            session: aiohttp 会话
+
+        Returns:
+            API 响应 JSON，包含 video id 和初始状态
+        """
+        url = f"{self.base_url}{self.CREATE_ENDPOINT}"
+        headers = {"Authorization": f"Bearer {self.api_key}"}
+
+        # 检查是否有图片
+        has_images = any([first_frame_bytes, last_frame_bytes, reference_bytes])
+
+        if has_images:
+            # 有图片：multipart/form-data + PNG 文件上传
+            if first_frame_bytes and len(first_frame_bytes) > self.max_request_size:
+                raise ValueError(f"首帧图片过大，超过 {self.max_request_size / 1024 / 1024:.0f}MB 限制")
+            if last_frame_bytes and len(last_frame_bytes) > self.max_request_size:
+                raise ValueError(f"尾帧图片过大，超过 {self.max_request_size / 1024 / 1024:.0f}MB 限制")
+            if reference_bytes and len(reference_bytes) > self.max_request_size:
+                raise ValueError(f"参考图片过大，超过 {self.max_request_size / 1024 / 1024:.0f}MB 限制")
+
+            form = aiohttp.FormData()
+            form.add_field("prompt", prompt)
+            form.add_field("model", model)
+            form.add_field("seconds", str(seconds))
+            form.add_field("size", size)
+            # 注意：seed 不被上游 API 接受，仅在 ComfyUI 节点侧用于缓存刷新
+            # if seed is not None:
+            #     form.add_field("seed", str(seed))
+
+            # 使用 input_reference 字段（OpenAI兼容格式）
+            # 尝试支持多张图片：按顺序添加多个 input_reference 字段
+            if first_frame_bytes:
+                form.add_field(
+                    "input_reference",
+                    first_frame_bytes,
+                    filename="first_frame.png",
+                    content_type="image/png",
+                )
+            if last_frame_bytes:
+                form.add_field(
+                    "input_reference",
+                    last_frame_bytes,
+                    filename="last_frame.png",
+                    content_type="image/png",
+                )
+            if reference_bytes:
+                form.add_field(
+                    "input_reference",
+                    reference_bytes,
+                    filename="reference.png",
+                    content_type="image/png",
+                )
+
+            send_kwargs: Dict[str, Any] = {"data": form, "headers": headers}
+        else:
+            # 无图片：application/json
+            body: Dict[str, Any] = {
+                "model": model,
+                "prompt": prompt,
+                "seconds": str(seconds),
+                "size": size,
+            }
+            # 注意：seed 不被上游 API 接受，仅在 ComfyUI 节点侧用于缓存刷新
+            # if seed is not None:
+            #     body["seed"] = str(seed)
+            send_kwargs = {"json": body, "headers": headers}
+
+        # 打印请求调试信息
+        import json
+        if has_images:
+            print(f"Veo: 使用 multipart/form-data 格式上传图片")
+        else:
+            print(f"Veo API 请求体: {json.dumps(body, ensure_ascii=False)}")
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            async with session.post(url, **send_kwargs) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    error_message = self._extract_error_message(error_text, response.status)
+                    raise RuntimeError(error_message)
+
+                resp_json = await response.json()
+                return resp_json
+
+        finally:
+            if close_session:
+                await session.close()
+
+    async def poll_video_status_async(
+        self,
+        video_id: str,
+        progress_callback: Optional[Callable[[int, float], None]] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> Dict[str, Any]:
+        """
+        轮询视频生成状态，直到完成或失败
+
+        Args:
+            video_id: 视频任务 ID
+            progress_callback: 进度回调 (progress_percent, elapsed_seconds)
+            session: aiohttp 会话
+
+        Returns:
+            最终状态的 API 响应
+
+        Raises:
+            RuntimeError: 生成失败
+        """
+        url = f"{self.base_url}{self.STATUS_ENDPOINT.format(video_id=video_id)}"
+        headers = self.get_headers(use_bearer_token=True)
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        interval = self.POLL_INITIAL_INTERVAL
+
+        try:
+            while True:
+                async with session.get(url, headers=headers) as response:
+                    if response.status != 200:
+                        error_text = await response.text()
+                        error_message = self._extract_error_message(error_text, response.status)
+                        raise RuntimeError(error_message)
+
+                    data = await response.json()
+
+                # status 兼容大小写
+                status = data.get("status", "").lower()
+
+                # progress 兼容整数和字符串
+                progress_raw = data.get("progress", 0)
+                if isinstance(progress_raw, str):
+                    try:
+                        progress = int(progress_raw.rstrip("%").strip())
+                    except ValueError:
+                        progress = 0
+                else:
+                    progress = int(progress_raw) if progress_raw else 0
+
+                if progress_callback:
+                    progress_callback(progress)
+
+                if status == "completed":
+                    return data
+
+                if status == "failed":
+                    error_info = data.get("error", {})
+                    error_msg = error_info.get("message", "未知错误") if isinstance(error_info, dict) else str(error_info)
+                    raise RuntimeError(f"视频生成失败: {error_msg}")
+
+                await asyncio.sleep(interval)
+                interval = min(interval * 1.5, self.POLL_MAX_INTERVAL)
+
+        finally:
+            if close_session:
+                await session.close()
+
+    async def download_video_async(
+        self,
+        video_id: str,
+        save_path: str,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> str:
+        """
+        下载生成的视频文件
+
+        Returns:
+            保存的文件路径
+        """
+        url = f"{self.base_url}{self.CONTENT_ENDPOINT.format(video_id=video_id)}"
+        headers = self.get_headers(use_bearer_token=True)
+
+        close_session = False
+        if session is None:
+            session = self._make_session()
+            close_session = True
+
+        try:
+            async with session.get(url, headers=headers, allow_redirects=True) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    error_message = self._extract_error_message(error_text, response.status)
+                    raise RuntimeError(f"视频下载失败: {error_message}")
+
+                content_type = response.headers.get("Content-Type", "")
+
+                if "application/json" in content_type:
+                    data = await response.json()
+                    download_url = data.get("url") or data.get("download_url")
+                    if not download_url:
+                        raise RuntimeError("视频下载失败: 响应中未找到下载链接")
+                    await self._download_from_url(download_url, save_path, session)
+                else:
+                    os.makedirs(os.path.dirname(save_path), exist_ok=True)
+                    with open(save_path, "wb") as f:
+                        async for chunk in response.content.iter_chunked(8192):
+                            f.write(chunk)
+
+            return save_path
+
+        finally:
+            if close_session:
+                await session.close()
+
+    # ------------------------------------------------------------------
+    # 同步包装
+    # ------------------------------------------------------------------
+
+    def generate_video_sync(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_path: str,
+        first_frame_bytes: Optional[bytes] = None,
+        last_frame_bytes: Optional[bytes] = None,
+        reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        progress_callback: Optional[Callable[[int], None]] = None,
+        on_stage: Optional[Callable[[str], None]] = None,
+    ) -> str:
+        """
+        同步执行完整的视频生成流程（创建 → 轮询 → 下载）
+        """
+
+        async def _run():
+            connector = aiohttp.TCPConnector(ssl=False, limit=0)
+            async with aiohttp.ClientSession(connector=connector) as session:
+                # 1. 提交任务
+                if on_stage:
+                    on_stage("submitting")
+                result = await self.create_video_async(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=size,
+                    first_frame_bytes=first_frame_bytes,
+                    last_frame_bytes=last_frame_bytes,
+                    reference_bytes=reference_bytes,
+                    seed=seed,
+                    session=session,
+                )
+                video_id = result.get("id")
+                if not video_id:
+                    raise RuntimeError("API 未返回视频任务 ID")
+
+                if on_stage:
+                    on_stage(f"submitted:{video_id}")
+
+                # 2. 轮询状态
+                if on_stage:
+                    on_stage("polling")
+                await self.poll_video_status_async(
+                    video_id=video_id,
+                    progress_callback=progress_callback,
+                    session=session,
+                )
+
+                # 3. 下载视频
+                if on_stage:
+                    on_stage("downloading")
+                path = await self.download_video_async(
+                    video_id=video_id,
+                    save_path=save_path,
+                    session=session,
+                )
+
+                if on_stage:
+                    on_stage("done")
+                return path
+
+        return self.run_async_in_thread(_run())
+
+    def generate_batch_videos_sync(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_paths: List[str],
+        first_frame_bytes: Optional[bytes] = None,
+        last_frame_bytes: Optional[bytes] = None,
+        reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        progress_callback: Optional[Callable[[int, int, bool, Optional[str]], None]] = None,
+    ) -> List[str]:
+        """
+        同步并发生成多个视频
+        """
+        async def _run():
+            batch_size = len(save_paths)
+            connector = aiohttp.TCPConnector(ssl=False, limit=0)
+
+            async def generate_one(save_path: str):
+                return await self._generate_one_video_async(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=size,
+                    save_path=save_path,
+                    first_frame_bytes=first_frame_bytes,
+                    last_frame_bytes=last_frame_bytes,
+                    reference_bytes=reference_bytes,
+                    seed=seed,
+                )
+
+            async with aiohttp.ClientSession(connector=connector) as session:
+                tasks = [generate_one(p) for p in save_paths]
+                results = await asyncio.gather(*tasks, return_exceptions=True)
+
+            completed = 0
+            paths: List[str] = []
+            first_error = None
+            for i, result in enumerate(results):
+                if isinstance(result, Exception):
+                    error_msg = str(result)
+                    print(f"Veo: 第 {i + 1} 个视频生成失败")
+                    if first_error is None:
+                        first_error = result
+                    if progress_callback:
+                        progress_callback(i + 1, batch_size, False, error_msg)
+                else:
+                    completed += 1
+                    paths.append(result)
+                    if progress_callback:
+                        progress_callback(completed, batch_size, True, None)
+
+            if not paths:
+                if first_error:
+                    raise first_error
+                raise RuntimeError(f"批量视频生成失败，{batch_size} 个任务全部失败")
+
+            return paths
+
+        return self.run_async_in_thread(_run())
+
+    async def _generate_one_video_async(
+        self,
+        prompt: str,
+        model: str,
+        seconds: int,
+        size: str,
+        save_path: str,
+        first_frame_bytes: Optional[bytes] = None,
+        last_frame_bytes: Optional[bytes] = None,
+        reference_bytes: Optional[bytes] = None,
+        seed: Optional[int] = None,
+        session: Optional[aiohttp.ClientSession] = None,
+    ) -> str:
+        """异步生成单个视频"""
+        result = await self.create_video_async(
+            prompt=prompt,
+            model=model,
+            seconds=seconds,
+            size=size,
+            first_frame_bytes=first_frame_bytes,
+            last_frame_bytes=last_frame_bytes,
+            reference_bytes=reference_bytes,
+            seed=seed,
+            session=session,
+        )
+        video_id = result.get("id")
+        if not video_id:
+            raise RuntimeError("API 未返回视频任务 ID")
+
+        await self.poll_video_status_async(video_id=video_id, session=session)
+        path = await self.download_video_async(
+            video_id=video_id, save_path=save_path, session=session
+        )
+        return path
+
+    # ------------------------------------------------------------------
+    # 内部辅助方法
+    # ------------------------------------------------------------------
+
+    async def _download_from_url(
+        self,
+        url: str,
+        save_path: str,
+        session: aiohttp.ClientSession,
+    ) -> None:
+        """从给定 URL 下载文件到本地路径"""
+        os.makedirs(os.path.dirname(save_path), exist_ok=True)
+        async with session.get(url) as response:
+            if response.status != 200:
+                raise RuntimeError(f"从下载链接获取视频失败 (状态码: {response.status})")
+            with open(save_path, "wb") as f:
+                async for chunk in response.content.iter_chunked(8192):
+                    f.write(chunk)
+
+    @staticmethod
+    def _extract_error_message(error_text: str, status_code: int) -> str:
+        """从错误响应中提取可读的错误信息"""
+        error_message = error_text
+        try:
+            error_json = json.loads(error_text)
+            if "error" in error_json:
+                if isinstance(error_json["error"], dict):
+                    error_message = error_json["error"].get("message", error_text)
+                else:
+                    error_message = str(error_json["error"])
+            elif "message" in error_json:
+                error_message = error_json["message"]
+        except (json.JSONDecodeError, KeyError):
+            pass
+
+        status_hints = {
+            400: "请求参数错误 (400)",
+            401: "认证失败 (401)，请检查 API 密钥",
+            403: "权限不足 (403)，请检查账户权限或余额",
+            429: "请求频率超限 (429)，请稍后重试",
+            503: "服务暂时不可用 (503)，请稍后重试",
+            504: "请求超时 (504)，请稍后重试",
+        }
+        hint = status_hints.get(status_code, f"API 请求失败 (状态码: {status_code})")
+        return f"{hint}\nAPI 返回: {error_message}"
@@ -0,0 +1,555 @@
+<!doctype html>
+<html lang="zh-CN">
+<head>
+  <meta charset="utf-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1" />
+  <title>O1Key 笔记侧栏草图</title>
+  <style>
+    :root {
+      --bg: #151515;
+      --rail: #1b1b1b;
+      --panel: #202020;
+      --panel-2: #262626;
+      --field: #181818;
+      --line: rgba(255,255,255,.09);
+      --line-2: rgba(255,255,255,.16);
+      --text: #e6e6e6;
+      --soft: #b5b5b5;
+      --muted: #777;
+      --blue: #4f8cff;
+      --blue-2: #7eb8f7;
+      --blue-soft: rgba(79,140,255,.14);
+      --danger: #d76565;
+    }
+
+    * { box-sizing: border-box; }
+    body {
+      margin: 0;
+      min-height: 100vh;
+      background: #101010;
+      color: var(--text);
+      font-family: "Microsoft YaHei", "Segoe UI", sans-serif;
+      display: grid;
+      place-items: center;
+    }
+
+    .frame {
+      width: 1180px;
+      height: 740px;
+      background: var(--bg);
+      border: 1px solid #303030;
+      display: grid;
+      grid-template-columns: 52px 352px 1fr;
+      overflow: hidden;
+      box-shadow: 0 18px 60px rgba(0,0,0,.45);
+      position: relative;
+    }
+
+    .rail {
+      background: var(--rail);
+      border-right: 1px solid var(--line);
+      padding: 10px 7px;
+      display: flex;
+      flex-direction: column;
+      gap: 8px;
+      align-items: center;
+    }
+
+    .rail button {
+      width: 36px;
+      height: 36px;
+      border: 1px solid transparent;
+      border-radius: 8px;
+      background: transparent;
+      color: #8a8a8a;
+      display: grid;
+      place-items: center;
+      cursor: default;
+      font-size: 15px;
+    }
+
+    .rail button.active {
+      color: var(--blue-2);
+      background: rgba(79,140,255,.13);
+      border-color: rgba(79,140,255,.36);
+    }
+
+    .note-panel {
+      background: var(--panel);
+      border-right: 1px solid var(--line);
+      display: flex;
+      flex-direction: column;
+      min-width: 0;
+      min-height: 0;
+    }
+
+    .header {
+      padding: 14px 14px 10px;
+      border-bottom: 1px solid rgba(255,255,255,.06);
+      flex: 0 0 auto;
+    }
+
+    .title-row,
+    .actions,
+    .tag-strip,
+    .note-top,
+    .note-meta,
+    .edit-top,
+    .edit-actions,
+    .tag-row {
+      display: flex;
+      align-items: center;
+    }
+
+    .title-row {
+      justify-content: space-between;
+      margin-bottom: 12px;
+    }
+
+    .title {
+      font-size: 14px;
+      font-weight: 700;
+      letter-spacing: .2px;
+    }
+
+    .actions { gap: 5px; }
+
+    .icon-btn {
+      width: 28px;
+      height: 28px;
+      border: 1px solid var(--line);
+      border-radius: 7px;
+      color: #9b9b9b;
+      background: rgba(255,255,255,.035);
+      display: grid;
+      place-items: center;
+    }
+
+    .icon-btn.primary {
+      color: white;
+      background: var(--blue);
+      border-color: transparent;
+    }
+
+    .search {
+      height: 34px;
+      border: 1px solid var(--line);
+      border-radius: 8px;
+      background: rgba(255,255,255,.045);
+      display: flex;
+      align-items: center;
+      gap: 8px;
+      padding: 0 10px;
+      color: var(--muted);
+      font-size: 12px;
+    }
+
+    .tag-strip {
+      gap: 6px;
+      margin-top: 10px;
+      overflow: hidden;
+    }
+
+    .tag-filter {
+      height: 26px;
+      padding: 0 9px;
+      border-radius: 7px;
+      border: 1px solid var(--line);
+      background: transparent;
+      color: #969696;
+      font-size: 12px;
+      white-space: nowrap;
+    }
+
+    .tag-filter.active {
+      color: var(--blue-2);
+      border-color: rgba(79,140,255,.36);
+      background: var(--blue-soft);
+    }
+
+    .list {
+      flex: 1;
+      min-height: 0;
+      overflow: hidden;
+      padding: 8px 10px 12px;
+      display: flex;
+      flex-direction: column;
+      gap: 6px;
+    }
+
+    .scroll-hint {
+      height: 24px;
+      color: #606060;
+      font-size: 11px;
+      display: flex;
+      align-items: center;
+      justify-content: center;
+      flex: 0 0 auto;
+      border-top: 1px solid rgba(255,255,255,.04);
+      margin: 2px 4px 0;
+    }
+
+    .note {
+      border: 1px solid transparent;
+      border-radius: 8px;
+      padding: 10px;
+      background: transparent;
+    }
+
+    .note.active {
+      background: rgba(255,255,255,.055);
+      border-color: rgba(255,255,255,.11);
+      box-shadow: inset 2px 0 0 var(--blue);
+    }
+
+    .note-title {
+      font-size: 13px;
+      color: #e0e0e0;
+      white-space: nowrap;
+      overflow: hidden;
+      text-overflow: ellipsis;
+      flex: 1;
+      min-width: 0;
+    }
+
+    .note-action {
+      color: #777;
+      font-size: 11px;
+      margin-left: 8px;
+    }
+
+    .note-text {
+      margin-top: 6px;
+      font-size: 12px;
+      line-height: 1.45;
+      color: #8c8c8c;
+      display: -webkit-box;
+      -webkit-line-clamp: 3;
+      -webkit-box-orient: vertical;
+      overflow: hidden;
+    }
+
+    .note-meta {
+      margin-top: 8px;
+      justify-content: space-between;
+      color: #666;
+      font-size: 10px;
+      gap: 8px;
+    }
+
+    .tags {
+      display: flex;
+      gap: 4px;
+      min-width: 0;
+      overflow: hidden;
+    }
+
+    .tag {
+      color: var(--blue-2);
+      background: rgba(79,140,255,.1);
+      border: 1px solid rgba(79,140,255,.2);
+      border-radius: 5px;
+      padding: 2px 5px;
+      white-space: nowrap;
+    }
+
+    .canvas {
+      position: relative;
+      background:
+        linear-gradient(rgba(255,255,255,.025) 1px, transparent 1px),
+        linear-gradient(90deg, rgba(255,255,255,.025) 1px, transparent 1px);
+      background-size: 26px 26px;
+      min-width: 0;
+      overflow: hidden;
+    }
+
+    .canvas-label {
+      position: absolute;
+      left: 24px;
+      top: 22px;
+      color: #5f5f5f;
+      font-size: 12px;
+    }
+
+    .node {
+      position: absolute;
+      width: 186px;
+      height: 92px;
+      border-radius: 8px;
+      border: 1px solid rgba(255,255,255,.12);
+      background: #222;
+      box-shadow: 0 12px 30px rgba(0,0,0,.2);
+    }
+
+    .node.one { left: 160px; top: 160px; }
+    .node.two { left: 430px; top: 290px; }
+    .node::before {
+      content: "";
+      display: block;
+      height: 28px;
+      border-bottom: 1px solid rgba(255,255,255,.08);
+      background: rgba(79,140,255,.12);
+      border-radius: 8px 8px 0 0;
+    }
+
+    .edit-popover {
+      position: absolute;
+      right: 24px;
+      top: 76px;
+      width: 420px;
+      height: 588px;
+      border: 1px solid rgba(79,140,255,.28);
+      border-radius: 10px;
+      background: #202020;
+      box-shadow: 0 22px 70px rgba(0,0,0,.46);
+      display: flex;
+      flex-direction: column;
+      overflow: hidden;
+    }
+
+    .edit-top {
+      height: 48px;
+      padding: 0 14px;
+      border-bottom: 1px solid rgba(255,255,255,.08);
+      justify-content: space-between;
+      flex: 0 0 auto;
+    }
+
+    .edit-title {
+      font-size: 13px;
+      font-weight: 700;
+    }
+
+    .edit-body {
+      padding: 12px;
+      display: flex;
+      flex-direction: column;
+      gap: 9px;
+      min-height: 0;
+      flex: 1;
+    }
+
+    .input,
+    .textarea {
+      border: 1px solid var(--line);
+      border-radius: 8px;
+      background: rgba(255,255,255,.045);
+      color: #ddd;
+      font-family: inherit;
+    }
+
+    .input {
+      height: 34px;
+      padding: 0 10px;
+      font-size: 13px;
+      font-weight: 700;
+      display: flex;
+      align-items: center;
+    }
+
+    .textarea {
+      flex: 1;
+      min-height: 220px;
+      padding: 12px;
+      font-size: 13px;
+      line-height: 1.6;
+      white-space: pre-wrap;
+      font-family: "JetBrains Mono", "Consolas", monospace;
+    }
+
+    .tag-editor {
+      min-height: 74px;
+      border: 1px solid var(--line);
+      border-radius: 8px;
+      background: var(--field);
+      padding: 8px;
+    }
+
+    .tag-row {
+      gap: 6px;
+      flex-wrap: wrap;
+    }
+
+    .tag-token {
+      height: 24px;
+      display: inline-flex;
+      align-items: center;
+      gap: 6px;
+      border-radius: 6px;
+      padding: 0 7px;
+      border: 1px solid rgba(79,140,255,.25);
+      color: var(--blue-2);
+      background: rgba(79,140,255,.1);
+      font-size: 12px;
+    }
+
+    .tag-token .x {
+      color: #8baee8;
+      font-size: 13px;
+    }
+
+    .tag-input {
+      height: 24px;
+      min-width: 106px;
+      padding: 0 6px;
+      border: 1px dashed rgba(255,255,255,.14);
+      border-radius: 6px;
+      color: #8f8f8f;
+      display: inline-flex;
+      align-items: center;
+      font-size: 12px;
+    }
+
+    .hint {
+      margin-top: 7px;
+      color: #666;
+      font-size: 11px;
+    }
+
+    .edit-actions {
+      gap: 8px;
+      padding: 12px;
+      border-top: 1px solid rgba(255,255,255,.08);
+      flex: 0 0 auto;
+    }
+
+    .btn {
+      height: 32px;
+      border-radius: 7px;
+      border: 1px solid var(--line);
+      background: rgba(255,255,255,.04);
+      color: #aaa;
+      padding: 0 12px;
+      font-size: 12px;
+    }
+
+    .btn.primary {
+      background: var(--blue);
+      border-color: transparent;
+      color: white;
+      font-weight: 700;
+      flex: 1;
+    }
+
+    .btn.cancel {
+      color: #ccc;
+    }
+
+    .btn.danger {
+      color: #ee9f9f;
+      border-color: rgba(215,101,101,.24);
+    }
+  </style>
+</head>
+<body>
+  <div class="frame">
+    <nav class="rail">
+      <button>☰</button>
+      <button>💬</button>
+      <button class="active">✎</button>
+      <button>🖼</button>
+      <button>⚙</button>
+    </nav>
+
+    <aside class="note-panel">
+      <div class="header">
+        <div class="title-row">
+          <div class="title">笔记</div>
+          <div class="actions">
+            <button class="icon-btn">⇩</button>
+            <button class="icon-btn primary">＋</button>
+          </div>
+        </div>
+        <div class="search">⌕ 搜索笔记内容或标签</div>
+        <div class="tag-strip">
+          <div class="tag-filter active">全部 18</div>
+          <div class="tag-filter">#产品图</div>
+          <div class="tag-filter">#Nano Banana</div>
+          <div class="tag-filter">#负向词</div>
+        </div>
+      </div>
+
+      <div class="list">
+        <div class="note active">
+          <div class="note-top">
+            <div class="note-title">产品主图：高级玻璃质感</div>
+            <div class="note-action">编辑</div>
+          </div>
+          <div class="note-text">Clean studio lighting, translucent glass material, subtle caustics, soft shadow, premium product photography...</div>
+          <div class="note-meta"><div class="tags"><span class="tag">#产品图</span><span class="tag">#玻璃</span></div><span>今天 14:22</span></div>
+        </div>
+        <div class="note">
+          <div class="note-top">
+            <div class="note-title">Nano Banana 参考图经验</div>
+            <div class="note-action">编辑</div>
+          </div>
+          <div class="note-text">参考图越多越容易跑偏，主体一致性优先用 1-3 张图；复杂场景建议分两步...</div>
+          <div class="note-meta"><div class="tags"><span class="tag">#Nano Banana</span><span class="tag">#参考图</span></div><span>昨天</span></div>
+        </div>
+        <div class="note">
+          <div class="note-top">
+            <div class="note-title">电商模特换装模板</div>
+            <div class="note-action">编辑</div>
+          </div>
+          <div class="note-text">Keep face identity, preserve pose, replace outfit with [服装描述], realistic fabric texture...</div>
+          <div class="note-meta"><div class="tags"><span class="tag">#电商</span><span class="tag">#换装</span></div><span>05/25</span></div>
+        </div>
+        <div class="note">
+          <div class="note-top">
+            <div class="note-title">常用负向词</div>
+            <div class="note-action">编辑</div>
+          </div>
+          <div class="note-text">low quality, blurry, deformed hands, extra fingers, bad anatomy, distorted text, watermark...</div>
+          <div class="note-meta"><div class="tags"><span class="tag">#负向词</span><span class="tag">#通用</span></div><span>05/20</span></div>
+        </div>
+        <div class="note">
+          <div class="note-top">
+            <div class="note-title">批量任务命名经验</div>
+            <div class="note-action">编辑</div>
+          </div>
+          <div class="note-text">小批量先跑 2-3 张确认风格，固定提示词和参考图后再放大批量数量...</div>
+          <div class="note-meta"><div class="tags"><span class="tag">#批量</span><span class="tag">#工作流</span></div><span>05/18</span></div>
+        </div>
+        <div class="scroll-hint">列表默认可滚动，编辑框不常驻</div>
+      </div>
+    </aside>
+
+    <main class="canvas">
+      <div class="canvas-label">ComfyUI 画布区域，笔记列表不再占用右侧空间</div>
+      <div class="node one"></div>
+      <div class="node two"></div>
+    </main>
+
+    <section class="edit-popover">
+      <div class="edit-top">
+        <div class="edit-title">编辑笔记</div>
+        <button class="icon-btn">×</button>
+      </div>
+      <div class="edit-body">
+        <div class="input">产品主图：高级玻璃质感</div>
+        <div class="tag-editor">
+          <div class="tag-row">
+            <span class="tag-token">产品图 <span class="x">×</span></span>
+            <span class="tag-token">玻璃 <span class="x">×</span></span>
+            <span class="tag-token">灯光 <span class="x">×</span></span>
+            <span class="tag-input">＋ 添加标签</span>
+          </div>
+          <div class="hint">只保留标签作为组织方式；可搜索、筛选、删除。</div>
+        </div>
+        <div class="textarea">Clean studio lighting, translucent glass material, subtle caustics, soft shadow, premium product photography, 85mm lens, minimal background, high detail.
+
+使用方式：
+1. 把产品图作为参考图输入
+2. 保留主体轮廓，只调整材质和灯光
+3. 如果玻璃过亮，降低 “caustics” 权重</div>
+      </div>
+      <div class="edit-actions">
+        <button class="btn cancel">取消</button>
+        <button class="btn">复制</button>
+        <button class="btn danger">删除</button>
+        <button class="btn primary">保存</button>
+      </div>
+    </section>
+  </div>
+</body>
+</html>
@@ -8,7 +8,7 @@
    3. 重新启用模型: 将模型的 enabled 字段改回 True

 模型类型:
-    - GEMINI_MODELS: Nano Banana Pro 图像生成模型
+    - GEMINI_MODELS: Nano Banana 图像生成模型
    - GEMINI_FLASH_MODELS: Google Gemini Flash 文本生成模型

 示例:
@@ -17,14 +17,21 @@
        "id": "gemini-新模型名称",
        "description": "模型说明和特点",
        "enabled": True,
-        "endpoint_type": "standard"  # 端点类型: "dynamic", "standard", "flatfee"
+        "endpoint_type": "standard",
+        "endpoint": "/v1beta/models/gemini-新模型名称:generateContent",
+        "thinking_config": {
+            "不思考": None,
+            "低": "low",
+            "中": None,
+            "高": "high"
+        }
    }
    
    临时关闭模型:
    将对应模型的 "enabled": True 改为 "enabled": False
 """

-from typing import List, Dict, Optional
+from typing import List, Dict, Optional, Tuple


 # ============================================================
@@ -32,45 +39,72 @@ from typing import List, Dict, Optional
 # ============================================================

 # ============================================================
-# Nano Banana Pro 图像生成模型
+# Nano Banana 图像生成模型
 # ============================================================

 GEMINI_MODELS = [
    {
-        "id": "nano-banana-pro",
-        "description": "Nano Banana Pro,根据分辨率自动选择端点 (1K/2K/4K),高性能图像生成模型",
+        "id": "nano-banana-pro-次卡",
+        "description": "Nano Banana Pro 次卡,根据分辨率自动选择端点 (1K/2K/4K),高性能图像生成模型",
        "enabled": True,
+        "provider": "gemini_async",
        "endpoint_type": "dynamic",
-        "endpoint": None  # 动态端点，由代码根据分辨率选择
+        "endpoint": None,  # 动态端点，由代码根据分辨率选择
+        "supported_aspect_ratios": [
+            "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9"
+        ],
+        "supported_resolutions": ["1K", "2K", "4K"]
    },
    {
-        "id": "gemini-3-pro-image-preview-url",
-        "description": "URL 模式,根据分辨率自动选择端点 (1K/2K/4K),推荐用于需要不同分辨率的场景",
-        "enabled": False,
-        "endpoint_type": "dynamic",
-        "endpoint": None  # 动态端点，由代码根据分辨率选择
-    },
-    {
-        "id": "gemini-3-pro-image-preview",
-        "description": "标准模式,固定端点,适用于常规图像生成",
+        "id": "nano-banana-pro-官方计费",
+        "description": "Nano Banana Pro 官方计费，按分辨率路由 (1K/2K/4K)，使用官方计费通道",
        "enabled": True,
-        "endpoint_type": "standard",
-        "endpoint": "/v1beta/models/gemini-3-pro-image-preview:generateContent"
+        "provider": "gemini_async",
+        "endpoint_type": "dynamic",
+        "endpoint": None,  # 动态端点，由代码根据分辨率选择
+        "supported_aspect_ratios": [
+            "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9"
+        ],
+        "supported_resolutions": ["1K", "2K", "4K"]
    },
    {
-        "id": "gemini-3-pro-image-preview-flatfee",
-        "description": "固定费用模式,固定端点,按固定价格计费 (暂时不可用-504错误)",
-        "enabled": False,  # 暂时禁用：端点返回 504 错误
-        "endpoint_type": "flatfee",
-        "endpoint": "/v1beta/models/gemini-3-pro-image-preview-flatfee:generateContent"
+        "id": "nano-banana-2-次卡",
+        "description": "Nano Banana 2 次卡，根据分辨率自动选择端点 (512px/1K/2K/4K)，图像生成模型",
+        "enabled": True,
+        "provider": "gemini_async",
+        "endpoint_type": "dynamic",
+        "endpoint": None,  # 动态端点，由代码根据分辨率选择
+        "supported_aspect_ratios": [
+            "1:1", "1:4", "1:8", "2:3", "3:2", "3:4", "4:1", "4:3", "4:5", "5:4",
+            "8:1", "9:16", "16:9", "21:9"
+        ],
+        "supported_resolutions": ["512px", "1K", "2K", "4K"]
    },
    {
-        "id": "nano-banana-2",
-        "description": "Nano Banana 2 模型,固定端点,适用于高质量图像生成",
-        "enabled": False,
+        "id": "nano-banana-2-官方计费",
+        "description": "Nano Banana 2 官方计费，按分辨率路由 (512/1K/2K/4K)，使用官方计费通道",
+        "enabled": True,
+        "provider": "gemini_async",
+        "endpoint_type": "dynamic",
+        "endpoint": None,  # 动态端点，由代码根据分辨率选择
+        "supported_aspect_ratios": [
+            "1:1", "1:4", "1:8", "2:3", "3:2", "3:4", "4:1", "4:3", "4:5", "5:4",
+            "8:1", "9:16", "16:9", "21:9"
+        ],
+        "supported_resolutions": ["512px", "1K", "2K", "4K"]
+    },
+    {
+        "id": "nano-banana-次卡",
+        "description": "Nano Banana 次卡，固定端点，图像生成模型",
+        "enabled": True,
+        "provider": "gemini_async",
        "endpoint_type": "standard",
-        "endpoint": "/v1beta/models/nano-banana-2:generateContent"
-    }
+        "endpoint": "/v1beta/models/nano-banana:generateContent",
+        "supported_aspect_ratios": [
+            "1:1", "2:3", "3:2", "3:4", "4:3", "4:5", "5:4", "9:16", "16:9", "21:9"
+        ],
+        "supported_resolutions": ["1K"]
+    },
 ]


@@ -80,14 +114,29 @@ GEMINI_MODELS = [

 GEMINI_FLASH_MODELS = [
    {
-        "id": "gemini-3-flash-preview",
-        "description": "Gemini 3 Flash,快速多模态文本生成,支持图片和视频输入",
+        "id": "gemini-3.5-flash",
+        "description": "Gemini 3.5 Flash,快速多模态文本生成,通过 thinkingConfig 控制思考等级",
        "enabled": True,
-        "endpoints": {
-            "不思考": "/v1beta/models/gemini-3-flash-preview-nothinking:generateContent",
-            "高": "/v1beta/models/gemini-3-flash-preview-high:generateContent"
+        "endpoint_type": "standard",
+        "endpoint": "/v1beta/models/gemini-3.5-flash:generateContent",
+        "thinking_config": {
+            "低": "low",
+            "中": "medium",
+            "高": "high"
        }
-    }
+    },
+
+    {
+        "id": "gemini-3.1-pro-preview",
+        "description": "Gemini 3.1 Pro,高性能多模态文本生成,通过 thinkingConfig 控制思考等级",
+        "enabled": True,
+        "endpoint_type": "standard",
+        "endpoint": "/v1beta/models/gemini-3.1-pro-preview:generateContent",
+        "thinking_config": {
+            "低": "low",
+            "中": "high"
+        }
+    },
 ]


@@ -184,6 +233,123 @@ def get_model_description(model_id: str) -> str:
    return config.get("description", "")


+def get_model_supported_aspect_ratios(model_id: str) -> List[str]:
+    """
+    获取模型支持的宽高比列表
+
+    Args:
+        model_id: 模型 ID
+
+    Returns:
+        支持的宽高比字符串列表，如果未配置则返回空列表
+
+    Example:
+        >>> get_model_supported_aspect_ratios("gemini-3-pro-image-preview")
+        ['1:1', '2:3', '3:2', ...]
+    """
+    config = get_model_config(model_id)
+    if config is None:
+        return []
+    return config.get("supported_aspect_ratios", [])
+
+
+def get_all_supported_aspect_ratios() -> List[str]:
+    """
+    获取所有启用模型支持的宽高比（去重合并）
+
+    Returns:
+        所有启用模型支持的宽高比列表（保持顺序、去重）
+
+    Example:
+        >>> get_all_supported_aspect_ratios()
+        ['1:1', '4:3', '3:4', '16:9', '9:16', '2:3', '3:2', '4:5', '5:4', '21:9', '1:4', '4:1', '1:8', '8:1']
+    """
+    seen = set()
+    result = []
+    for model in GEMINI_MODELS:
+        if not model.get("enabled", False):
+            continue
+        for ratio in model.get("supported_aspect_ratios", []):
+            if ratio not in seen:
+                seen.add(ratio)
+                result.append(ratio)
+    return result
+
+
+def get_model_supported_resolutions(model_id: str) -> List[str]:
+    """
+    获取模型支持的分辨率列表
+
+    Args:
+        model_id: 模型 ID
+
+    Returns:
+        支持的分辨率字符串列表，如果未配置则返回空列表
+
+    Example:
+        >>> get_model_supported_resolutions("gemini-3.1-flash-image-preview")
+        ['512', '1K', '2K', '4K']
+        >>> get_model_supported_resolutions("gemini-3-pro-image-preview")
+        ['1K', '2K', '4K']
+    """
+    config = get_model_config(model_id)
+    if config is None:
+        return []
+    return config.get("supported_resolutions", [])
+
+
+def get_all_supported_resolutions() -> List[str]:
+    """
+    获取所有启用模型支持的分辨率（去重合并，按从小到大固定顺序排列）
+
+    Returns:
+        所有启用模型支持的分辨率列表（按 512 → 1K → 2K → 4K 顺序）
+
+    Example:
+        >>> get_all_supported_resolutions()
+        ['512', '1K', '2K', '4K']
+    """
+    _ORDER = ["512px", "1K", "2K", "4K"]
+
+    seen = set()
+    for model in GEMINI_MODELS:
+        if not model.get("enabled", False):
+            continue
+        for res in model.get("supported_resolutions", []):
+            seen.add(res)
+
+    return [res for res in _ORDER if res in seen]
+
+
+def get_model_provider(model_id: str) -> Optional[str]:
+    """
+    获取模型的异步 Provider 名称
+
+    Args:
+        model_id: 模型 ID
+
+    Returns:
+        Provider 名称（如 "gemini_async"），如果模型未配置 provider 则返回 None
+    """
+    config = get_model_config(model_id)
+    if config is None:
+        return None
+    return config.get("provider")
+
+
+def get_enabled_async_models() -> List[str]:
+    """
+    获取所有启用的、支持异步模式的模型 ID 列表
+
+    Returns:
+        模型 ID 列表（仅包含配置了 provider 且 enabled 的模型）
+    """
+    return [
+        model["id"] for model in GEMINI_MODELS
+        if model.get("enabled", False) and model.get("provider")
+    ]
+
+
 def get_endpoint_type(model_id: str) -> Optional[str]:
    """
    获取模型的端点类型
@@ -226,6 +392,233 @@ def get_model_endpoint(model_id: str) -> Optional[str]:
    return config.get("endpoint")


+# ============================================================
+# Gemini Flash 模型工具函数
+# ============================================================
+
+# ============================================================
+# Sora 视频生成模型
+# ============================================================
+
+SORA_MODELS = [
+    {
+        "id": "sora-2",
+        "description": "Sora 2 官方模型，支持标准时长和分辨率",
+        "enabled": True,
+        "supported_seconds": [4, 8, 10, 12, 15],
+        "supported_sizes": ["720x1280", "1280x720"],
+        "seconds_category": "官方",  # 用于界面显示标签
+    },
+    {
+        "id": "sora-2-pro",
+        "description": "Sora 2 Pro 增强模型，支持扩展时长和竖屏/横屏高清分辨率",
+        "enabled": True,
+        "supported_seconds": [4, 8, 12, 15, 25],
+        "supported_sizes": ["720x1280", "1280x720", "1024x1792", "1792x1024"],
+        "seconds_category": "扩展",  # Pro 模型支持全部时长
+    },
+]
+
+# 秒数显示标签配置（用于界面下拉菜单）
+# key: 实际秒数, value: 显示文本
+SECONDS_DISPLAY_MAP = {
+    4: "4",
+    8: "8",
+    12: "12",
+    10: "10",
+    15: "15",
+    25: "25(pro)",
+}
+
+# 分辨率显示标签配置
+# key: 实际分辨率, value: (显示P数, 显示方向)
+RESOLUTION_DISPLAY_MAP = {
+    "720x1280": ("720P", "竖屏"),
+    "1280x720": ("720P", "横屏"),
+    "1024x1792": ("1080P", "竖屏"),
+    "1792x1024": ("1080P", "横屏"),
+}
+
+
+# ============================================================
+# Sora 模型工具函数
+# ============================================================
+
+def get_enabled_sora_models() -> List[str]:
+    """获取所有启用的 Sora 模型 ID 列表"""
+    return [model["id"] for model in SORA_MODELS if model.get("enabled", False)]
+
+
+def get_sora_model_config(model_id: str) -> Optional[Dict]:
+    """根据模型 ID 获取 Sora 模型的完整配置"""
+    for model in SORA_MODELS:
+        if model["id"] == model_id:
+            return model
+    return None
+
+
+def get_sora_supported_seconds(model_id: str) -> List[int]:
+    """获取 Sora 模型支持的视频时长列表（秒）"""
+    config = get_sora_model_config(model_id)
+    if config is None:
+        return []
+    return config.get("supported_seconds", [])
+
+
+def get_sora_supported_sizes(model_id: str) -> List[str]:
+    """获取 Sora 模型支持的分辨率列表"""
+    config = get_sora_model_config(model_id)
+    if config is None:
+        return []
+    return config.get("supported_sizes", [])
+
+
+def get_all_sora_seconds() -> List[int]:
+    """获取所有启用 Sora 模型支持的时长（去重、升序）"""
+    seen = set()
+    for model in SORA_MODELS:
+        if not model.get("enabled", False):
+            continue
+        for s in model.get("supported_seconds", []):
+            seen.add(s)
+    return sorted(seen)
+
+
+def get_all_sora_sizes() -> List[str]:
+    """获取所有启用 Sora 模型支持的分辨率（去重、保持顺序）"""
+    seen = set()
+    result = []
+    for model in SORA_MODELS:
+        if not model.get("enabled", False):
+            continue
+        for size in model.get("supported_sizes", []):
+            if size not in seen:
+                seen.add(size)
+                result.append(size)
+    return result
+
+
+def get_sora_seconds_with_labels(model_id: str) -> List[Tuple[str, int]]:
+    """
+    获取指定模型支持的秒数列表（带标签显示）
+    
+    Returns:
+        列表项为 (显示文本, 实际秒数)，如 [("4（官方）", 4), ("10（特殊）", 10)]
+    """
+    config = get_sora_model_config(model_id)
+    if config is None:
+        return []
+    
+    seconds_list = config.get("supported_seconds", [])
+    result = []
+    for s in seconds_list:
+        category = SECONDS_CATEGORIES.get(s, "")
+        label = f"{s}（{category}）" if category else str(s)
+        result.append((label, s))
+    return result
+
+
+def get_sora_sizes_with_labels(model_id: str) -> List[Tuple[str, str]]:
+    """
+    获取指定模型支持的分辨率列表（带独占标识）
+    
+    Returns:
+        列表项为 (显示文本, 实际分辨率)，如 [("720P 9:16 (720x1280)", "720x1280")]
+    """
+    from math import gcd
+    
+    config = get_sora_model_config(model_id)
+    if config is None:
+        return []
+    
+    sizes = config.get("supported_sizes", [])
+    result = []
+    
+    # 检查哪些分辨率是独占的（仅该模型支持）
+    all_sizes_count = {}
+    for m in SORA_MODELS:
+        if not m.get("enabled", False):
+            continue
+        for size in m.get("supported_sizes", []):
+            all_sizes_count[size] = all_sizes_count.get(size, 0) + 1
+    
+    for size in sizes:
+        # 解析分辨率
+        parts = size.lower().split("x")
+        w, h = int(parts[0]), int(parts[1])
+        short_side = min(w, h)
+        
+        # 分辨率等级
+        if short_side >= 1792:
+            res = "2K+"
+        elif short_side >= 1080:
+            res = "1K+"
+        elif short_side >= 720:
+            res = "720P"
+        else:
+            res = f"{short_side}P"
+        
+        # 比例
+        g = gcd(w, h)
+        ratio = f"{w // g}:{h // g}"
+        
+        # 检查是否独占
+        exclusive = all_sizes_count.get(size, 0) == 1
+        exclusive_tag = " [Pro独占]" if exclusive else ""
+        
+        # 方向
+        orientation = "竖屏" if h > w else "横屏" if w > h else "方形"
+        
+        label = f"{res} {ratio} {orientation}{exclusive_tag} ({size})"
+        result.append((label, size))
+    
+    return result
+
+
+# ============================================================
+# Google Veo 视频生成模型
+# ============================================================
+
+VEO_MODELS = [
+    {
+        "id": "Veo3.1",
+        "description": "Google Veo 3.1 视频生成模型，支持文生视频和图生视频",
+        "enabled": True,
+    },
+]
+
+# Veo 分辨率映射表
+# key: "分辨率_宽高比", value: 实际分辨率字符串
+VEO_RESOLUTION_MAP = {
+    # 720p
+    "720p_9:16": "720x1280",
+    "720p_16:9": "1280x720",
+    # 1080p
+    "1080p_9:16": "1080x1920",
+    "1080p_16:9": "1920x1080",
+    # 4K
+    "4K_9:16": "2160x3840",
+    "4K_16:9": "3840x2160",
+}
+
+
+# ============================================================
+# Veo 模型工具函数
+# ============================================================
+
+def get_enabled_veo_models() -> List[str]:
+    """获取所有启用的 Veo 模型 ID 列表"""
+    return [model["id"] for model in VEO_MODELS if model.get("enabled", False)]
+
+
+def get_veo_model_config(model_id: str) -> Optional[Dict]:
+    """根据模型 ID 获取 Veo 模型的完整配置"""
+    for model in VEO_MODELS:
+        if model["id"] == model_id:
+            return model
+    return None
+
+
 # ============================================================
 # Gemini Flash 模型工具函数
 # ============================================================
@@ -291,29 +684,24 @@ def is_flash_model_enabled(model_id: str) -> bool:
    return config.get("enabled", False)


-def get_flash_model_endpoint(model_id: str, thinking_depth: str = "不思考") -> Optional[str]:
+def get_flash_model_endpoint(model_id: str) -> Optional[str]:
    """
    获取 Flash 模型的 API 端点
    
    Args:
        model_id: 模型 ID
-        thinking_depth: 思考深度 ("不思考" 或 "高")
    
    Returns:
        API 端点路径，如果未找到则返回 None
    
    Example:
-        >>> get_flash_model_endpoint("gemini-3-flash-preview", "不思考")
-        '/v1beta/models/gemini-3-flash-preview-nothinking:generateContent'
-        >>> get_flash_model_endpoint("gemini-3-flash-preview", "高")
-        '/v1beta/models/gemini-3-flash-preview-high:generateContent'
+        >>> get_flash_model_endpoint("gemini-3-flash-preview")
+        '/v1beta/models/gemini-3-flash-preview:generateContent'
    """
    config = get_flash_model_config(model_id)
    if config is None:
        return None
-    
-    endpoints = config.get("endpoints", {})
-    return endpoints.get(thinking_depth)
+    return config.get("endpoint")


 def get_flash_model_description(model_id: str) -> str:
@@ -332,6 +720,83 @@ def get_flash_model_description(model_id: str) -> str:
    return config.get("description", "")


+def get_flash_model_thinking_level_value(model_id: str, thinking_level: str) -> Optional[str]:
+    """
+    获取指定模型在给定思考等级下应传入请求体的 thinkingLevel 值。
+    
+    仅对 endpoint_type="standard" 且配置了 thinking_config 的模型有效。
+    返回 None 表示该等级不受支持，请求体中不应包含 thinkingConfig。
+    
+    Args:
+        model_id: 模型 ID
+        thinking_level: 思考等级中文名（不思考/低/中/高）
+    
+    Returns:
+        API thinkingLevel 值（如 "low"/"medium"/"high"），或 None（不传参）
+    
+    Example:
+        >>> get_flash_model_thinking_level_value("gemini-3-pro-preview", "低")
+        'low'
+        >>> get_flash_model_thinking_level_value("gemini-3-pro-preview", "中")
+        None  # 不受支持，省略 thinkingConfig
+    """
+    config = get_flash_model_config(model_id)
+    if config is None:
+        return None
+    thinking_config = config.get("thinking_config")
+    if not thinking_config:
+        return None
+    return thinking_config.get(thinking_level)
+
+
+# 已弃用：动态端点模式下不再需要这些函数
+# def get_flash_model_thinking_levels(model_id: str) -> List[str]:
+#     """
+#     获取 Flash 模型支持的思考等级列表
+#     
+#     Args:
+#         model_id: 模型 ID
+#     
+#     Returns:
+#         思考等级列表（中文），如果未找到则返回空列表
+#     
+#     Example:
+#         >>> get_flash_model_thinking_levels("gemini-3-flash-preview")
+#         ['默认', '最低', '低', '中', '高']
+#     """
+#     config = get_flash_model_config(model_id)
+#     if config is None:
+#         return []
+#     
+#     thinking_levels = config.get("thinking_levels", {})
+#     return list(thinking_levels.keys())
+
+
+# def get_thinking_level_value(model_id: str, thinking_level: str) -> Optional[str]:
+#     """
+#     获取思考等级对应的 API 参数值
+#     
+#     Args:
+#         model_id: 模型 ID
+#         thinking_level: 思考等级（中文）
+#     
+#     Returns:
+#         API 参数值（英文），如果未找到则返回 None
+#     
+#     Example:
+#         >>> get_thinking_level_value("gemini-3-flash-preview", "默认")
+#         'high'
+#         >>> get_thinking_level_value("gemini-3-flash-preview", "最低")
+#         'minimal'
+#     """
+#     config = get_flash_model_config(model_id)
+#     if config is None:
+#         return None
+#     
+#     thinking_levels = config.get("thinking_levels", {})
+#     return thinking_levels.get(thinking_level)
+
+
 # ============================================================
 # 向后兼容性检查
 # ============================================================
@@ -392,8 +857,8 @@ def validate_flash_models_config() -> None:
    验证 Flash 模型配置的完整性
    
    检查:
-    - 每个模型必须有 id, description, enabled, endpoints 字段
-    - endpoints 必须包含所有思考深度选项
+    - 每个模型必须有 id, description, enabled 字段
+    - 每个模型必须有 endpoint 字段且格式正确
    - 至少有一个模型是启用的
    
    Raises:
@@ -402,8 +867,7 @@ def validate_flash_models_config() -> None:
    if not GEMINI_FLASH_MODELS:
        raise ValueError("GEMINI_FLASH_MODELS 列表不能为空")
    
-    required_fields = ["id", "description", "enabled", "endpoints"]
-    required_thinking_depths = ["不思考", "高"]
+    required_fields = ["id", "description", "enabled"]
    
    for i, model in enumerate(GEMINI_FLASH_MODELS):
        # 检查必需字段
@@ -411,24 +875,16 @@ def validate_flash_models_config() -> None:
            if field not in model:
                raise ValueError(f"Flash 模型 #{i} 缺少必需字段: {field}")
        
-        # 检查 endpoints 字典
-        endpoints = model.get("endpoints", {})
-        if not isinstance(endpoints, dict):
-            raise ValueError(f"Flash 模型 {model['id']} 的 endpoints 必须是字典")
+        # 检查端点配置
+        if "endpoint" not in model:
+            raise ValueError(f"Flash 模型 {model['id']} 缺少 'endpoint' 字段")
        
-        # 检查所有思考深度选项都有对应端点
-        for depth in required_thinking_depths:
-            if depth not in endpoints:
-                raise ValueError(
-                    f"Flash 模型 {model['id']} 的 endpoints 缺少 '{depth}' 思考深度"
-                )
-            
-            endpoint = endpoints[depth]
-            if not endpoint or not endpoint.startswith("/v1beta/models/"):
-                raise ValueError(
-                    f"Flash 模型 {model['id']} 的端点 '{endpoint}' 格式不正确。"
-                    f"应以 '/v1beta/models/' 开头"
-                )
+        endpoint = model.get("endpoint", "")
+        if not endpoint or not endpoint.startswith("/v1beta/models/"):
+            raise ValueError(
+                f"Flash 模型 {model['id']} 的 endpoint '{endpoint}' 格式不正确。"
+                f"应以 '/v1beta/models/' 开头"
+            )
    
    # 检查至少有一个启用的模型
    if not get_enabled_flash_models():
@@ -0,0 +1,364 @@
+"""
+K3 动作控制 自研节点
+用参考视频驱动参考图中人物动作，生成视频。
+视频通过 R2 上传后传 URL，图片转 base64 直传。
+"""
+
+import asyncio
+import io
+import json
+import os
+import struct
+import tempfile
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_async_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.r2_uploader import upload_video, upload_image
+from ..utils.image_utils import tensor_to_pil
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+try:
+    from comfy_api.latest import InputImpl
+    import folder_paths
+    _FOLDER_PATHS_OK = True
+except Exception:
+    _FOLDER_PATHS_OK = False
+
+
+# ── 常量 ──────────────────────────────────────────────────────────────────────
+
+_ENDPOINT_CREATE = "/kling/v1/videos/motion-control"
+_ENDPOINT_STATUS = "/kling/v1/videos/motion-control/{task_id}"
+
+_POLL_INIT = 5
+_POLL_MAX  = 15
+
+
+# ── 工具函数 ───────────────────────────────────────────────────────────────────
+
+
+
+# ── 视频时长检测（纯标准库，跨平台） ──────────────────────────────────────────
+
+def _parse_video_duration(data: bytes) -> float | None:
+    """从 MP4/MOV 原始字节解析时长（秒）。读取 mvhd box。"""
+    idx = data.find(b"mvhd")
+    if idx == -1:
+        return None
+    box = data[idx + 4:]
+    if len(box) < 32:
+        return None
+    version = box[0]
+    try:
+        if version == 0:
+            timescale = struct.unpack(">I", box[12:16])[0]
+            duration  = struct.unpack(">I", box[16:20])[0]
+        else:  # version == 1
+            timescale = struct.unpack(">I", box[20:24])[0]
+            duration  = struct.unpack(">Q", box[24:32])[0]
+    except struct.error:
+        return None
+    return (duration / timescale) if timescale > 0 else None
+
+
+def _get_video_duration(reference_video) -> float | None:
+    """从 ComfyUI VIDEO 对象获取视频时长（秒），失败返回 None。"""
+    try:
+        source = reference_video.get_stream_source()
+        if isinstance(source, str) and os.path.isfile(source):
+            with open(source, "rb") as f:
+                data = f.read()
+        elif isinstance(source, io.BytesIO):
+            source.seek(0)
+            data = source.read()
+        else:
+            return None
+        return _parse_video_duration(data)
+    except Exception:
+        return None
+
+
+def _validate_video_duration(reference_video, character_orientation: str):
+    """校验视频时长，超限时抛出 ValueError。解析失败时静默跳过。"""
+    duration = _get_video_duration(reference_video)
+    if duration is None:
+        print("[K3 动作控制] 无法解析视频时长，跳过校验。")
+        return
+    limit = 10 if character_orientation == "image" else 30
+    print(f"[K3 动作控制] 检测到视频时长: {duration:.2f}s（限制: 3~{limit}s）")
+    if not (3 <= duration <= limit):
+        orientation_label = "图片" if character_orientation == "image" else "视频"
+        raise ValueError(
+            f"参考视频时长 {duration:.1f}s 不符合要求。\n"
+            f"角色朝向为「{orientation_label}」时，时长须在 3~{limit}s 之间。"
+        )
+
+
+# ── 节点 ──────────────────────────────────────────────────────────────────────
+
+class K3MotionControl:
+    """K3 动作控制 自研 —— 用参考视频驱动参考图人物动作"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "参考图片":   ("IMAGE",),
+                "参考视频":   ("VIDEO",),
+                "提示词":     ("STRING", {"multiline": True, "default": ""}),
+                "网络线路":   (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型":       (["v3", "v2-6"], {"default": "v3"}),
+                "模式":       (["720p", "1080p"], {"default": "1080p"}),
+                "时长":       ([5, 10, 15, 20, 25, 30], {"default": 5}),
+                "角色朝向":   (["图片", "视频"], {"default": "图片"}),
+                "保留原声":   (["打开", "关闭"], {"default": "打开"}),
+                "seed": ("INT", {
+                    "default": 0, "min": 0, "max": 2147483647,
+                    "tooltip": "seed 仅控制节点是否重新运行，结果本身不可复现。",
+                }),
+            },
+        }
+
+    RETURN_TYPES  = ("VIDEO",)
+    RETURN_NAMES  = ("视频",)
+    FUNCTION      = "generate"
+    CATEGORY      = "comfyui_o1key/KVideo"
+
+    async def generate(self, 参考图片, 参考视频, 提示词, 保留原声, 角色朝向, 模式, 模型, 时长, 网络线路, seed, **kwargs):
+        api_key  = get_api_key_or_raise()
+        base_url = get_base_url_by_route(网络线路)
+        headers  = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type":  "application/json",
+        }
+
+        # ── 参数映射 ──────────────────────────────────────────────────
+        mode_api              = "std" if 模式 == "720p" else "pro"
+        model_name            = f"kling-{模型}-motion-{mode_api}-{时长}s"
+        character_orientation = "image" if 角色朝向 == "图片" else "video"
+        keep_sound            = "yes" if 保留原声 == "打开" else "no"
+        prompt                = 提示词.strip()
+
+        if len(prompt) > 2500:
+            raise ValueError(f"提示词长度（{len(prompt)}）超过上限 2500 个字符，请缩短后重试。")
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def _stage(s: str):
+            if s == "uploading":
+                print("[K3 动作控制] 上传视频到 R2...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif s == "submitting":
+                print("[K3 动作控制] 提交任务...")
+                if pbar: pbar.update_absolute(10, 100)
+            elif s.startswith("submitted:"):
+                print(f"[K3 动作控制] 任务已提交 → {s.split(':', 1)[1]}")
+                if pbar: pbar.update_absolute(15, 100)
+            elif s == "downloading":
+                print("[K3 动作控制] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif s == "done":
+                print("[K3 动作控制] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def _progress(pct: int):
+            if pbar: pbar.update_absolute(15 + int(pct * 0.84), 100)
+
+        # ── 视频时长校验 ──────────────────────────────────────────────
+        _validate_video_duration(参考视频, character_orientation)
+
+        # 参考视频时长不得超过所选时长（防止用长视频生成短计费）
+        _dur = _get_video_duration(参考视频)
+        if _dur is not None and _dur > 时长 + 0.5:
+            raise ValueError(
+                f"参考视频时长 {_dur:.1f}s 超过所选时长 {时长}s。\n"
+                f"请将时长调整为 ≥{_dur:.0f}s 的档位，或更换更短的参考视频。"
+            )
+
+        # ── 图片 & 视频上传 R2 → 获取公网 URL ────────────────────────
+        _stage("uploading")
+        check_interrupt()
+        pil_list  = tensor_to_pil(参考图片)
+        image_url = await upload_image(pil_list[0].convert("RGB"))
+        check_interrupt()
+        video_url = await upload_video(参考视频)
+
+        # ── 构建请求体 ────────────────────────────────────────────────
+        body: dict = {
+            "model_name":            model_name,
+            "model":                 model_name,
+            "image_url":             image_url,
+            "video_url":             video_url,
+            "character_orientation": character_orientation,
+            "mode":                  mode_api,
+            "keep_original_sound":   keep_sound,
+        }
+        if prompt:
+            body["prompt"] = prompt
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        tmp_fd, save_path = tempfile.mkstemp(suffix=".mp4", prefix="k3_motion_")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交任务
+            check_interrupt()
+            _stage("submitting")
+            create_url = f"{base_url}{_ENDPOINT_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url,
+                data=json.dumps(body, ensure_ascii=False).encode("utf-8"),
+                headers=headers, prefix="K3 动作控制提交: "
+            ))
+            check_interrupt()
+            text = await resp.text()
+            create_resp = json.loads(text)
+
+            # task_id 兼容扁平结构和 data 嵌套结构
+            task_id = (
+                create_resp.get("task_id")
+                or create_resp.get("id")
+                or create_resp.get("data", {}).get("task_id")
+            )
+            if not task_id:
+                raise RuntimeError(f"API 未返回任务 ID，响应：{create_resp}")
+            _stage(f"submitted:{task_id}")
+
+            # 2. 轮询
+            status_url = f"{base_url}{_ENDPOINT_STATUS.format(task_id=task_id)}"
+            interval   = _POLL_INIT
+            video_result_url = None
+
+            while True:
+                await interruptible_sleep(interval)
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    text = await resp.text()
+                    if resp.status != 200:
+                        try:
+                            err = json.loads(text)
+                            msg = err.get("message") or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                    sr = json.loads(text)
+
+                # 兼容扁平结构和 data 嵌套结构
+                data   = sr.get("data", sr)
+                status = extract_status(sr)
+
+                pct = extract_progress(sr)
+                print(f"[K3 动作控制] 生成中 {pct}%")
+                _progress(pct)
+
+                if is_success_status(status):
+                    video_result_url = extract_video_url(sr)
+                    break
+                elif is_failure_status(status, sr):
+                    err_msg = extract_error_message(sr)
+                    raise RuntimeError(f"K3 动作控制生成失败：{err_msg}")
+
+                interval = min(interval * 1.3, _POLL_MAX)
+
+            if not video_result_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{sr}")
+
+            # 3. 下载视频
+            check_interrupt()
+            _stage("downloading")
+            async with session.get(video_result_url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                os.close(tmp_fd)
+                with open(save_path, "wb") as f:
+                    async for chunk in resp.content.iter_chunked(8192):
+                        check_interrupt()
+                        f.write(chunk)
+
+        _stage("done")
+
+        if _FOLDER_PATHS_OK:
+            return (InputImpl.VideoFromFile(save_path),)
+        return (save_path,)
+
+
+# ── 视频时长检测测试节点 ──────────────────────────────────────────────────────
+
+class K3MotionVideoCheck:
+    """检测视频时长并校验是否满足动作控制的限制，不调用 API。"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "参考视频": ("VIDEO",),
+                "角色朝向": (["图片", "视频"], {"default": "图片"}),
+            },
+        }
+
+    RETURN_TYPES  = ("STRING",)
+    RETURN_NAMES  = ("检测结果",)
+    FUNCTION      = "check"
+    CATEGORY      = "comfyui_o1key/KVideo"
+    OUTPUT_NODE   = True
+
+    def check(self, 参考视频, 角色朝向):
+        character_orientation = "image" if 角色朝向 == "图片" else "video"
+        duration = _get_video_duration(参考视频)
+
+        if duration is None:
+            result = "❌ 无法解析视频时长（格式不支持或文件损坏）"
+            print(f"[K3 视频检测] {result}")
+            return (result,)
+
+        limit = 10 if character_orientation == "image" else 30
+        orientation_label = 角色朝向
+        ok = 3 <= duration <= limit
+
+        if ok:
+            result = (
+                f"✅ 时长检测通过\n"
+                f"视频时长: {duration:.2f}s\n"
+                f"角色朝向: {orientation_label}（限制 3~{limit}s）"
+            )
+        else:
+            result = (
+                f"❌ 时长检测不通过\n"
+                f"视频时长: {duration:.2f}s\n"
+                f"角色朝向: {orientation_label}（限制 3~{limit}s）\n"
+                f"请更换时长在 3~{limit}s 之间的视频。"
+            )
+
+        print(f"[K3 视频检测] {result}")
+        return (result,)
+
+
+# ── 节点注册 ──────────────────────────────────────────────────────────────────
+
+NODE_CLASS_MAPPINGS = {
+    "K3MotionControl":     K3MotionControl,
+    "K3MotionVideoCheck":  K3MotionVideoCheck,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "K3MotionControl":    "动作控制 K3 自研",
+    "K3MotionVideoCheck": "视频时长检测 K3",
+}
@@ -0,0 +1,344 @@
+"""
+K3 图生视频 自研节点（图生视频 / 多镜头）
+模型名根据 模式/时长/音频 动态拼接，不暴露在前端。
+起始帧为必填，仅作图生视频；多镜头功能待实现。
+"""
+
+import asyncio
+import json
+import os
+import tempfile
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_async_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+try:
+    from comfy_api.latest import InputImpl
+    import folder_paths
+    _FOLDER_PATHS_OK = True
+except Exception:
+    _FOLDER_PATHS_OK = False
+
+
+# ── 常量 ──────────────────────────────────────────────────────────────────────
+
+_MODEL_BASE = "kling-v3"          # 动态拼接为 kling-v3-{模式}-{时长}s-{voice}
+_MODES      = ["720p", "1080p", "4K"]
+_MODE_MAP   = {"720p": "std", "1080p": "pro", "4K": "4k"}
+
+_MULTI_SHOT_OPTIONS = [
+    "禁用",
+    "1个故事板",
+    "2个故事板",
+    "3个故事板",
+    "4个故事板",
+    "5个故事板",
+    "6个故事板",
+]
+
+_ENDPOINT_CREATE = "/v1/video/generations"
+_ENDPOINT_STATUS = "/v1/video/generations/{task_id}"
+
+_POLL_INIT = 3
+_POLL_MAX  = 15
+
+
+# ── 工具函数 ───────────────────────────────────────────────────────────────────
+
+def _image_to_base64(tensor) -> str:
+    pil = tensor_to_pil(tensor)
+    return encode_image_to_base64(pil[0], format="PNG")
+
+
+def _prepare_image_base64(tensor) -> str:
+    """转换并校验图片，不符合约束时自动等比缩放后返回 base64。"""
+    import io
+    import base64
+
+    pil_list = tensor_to_pil(tensor)
+    img = pil_list[0].convert("RGB")
+    w, h = img.size
+
+    # 1. 宽高比校验（无法通过等比缩放修复，直接报错）
+    ratio = w / h
+    if ratio < 1 / 2.5 or ratio > 2.5:
+        raise RuntimeError(
+            f"图片宽高比 {w}:{h}（{ratio:.2f}）超出允许范围 1:2.5 ~ 2.5:1，请裁剪后重试。"
+        )
+
+    # 2. 最小尺寸：任意边 < 300px 时等比放大
+    if w < 300 or h < 300:
+        scale = max(300 / w, 300 / h)
+        img = img.resize((int(w * scale), int(h * scale)), resample=1)  # LANCZOS=1
+
+    # 3. 文件大小：循环等比缩小直到 ≤ 10MB
+    MAX_BYTES = 10 * 1024 * 1024
+    for _ in range(20):                      # 最多迭代 20 次，防止死循环
+        buf = io.BytesIO()
+        img.save(buf, format="PNG")
+        if buf.tell() <= MAX_BYTES:
+            break
+        scale = (MAX_BYTES / buf.tell()) ** 0.5 * 0.95   # 留 5% 余量
+        new_w = int(img.width * scale)
+        new_h = int(img.height * scale)
+        if new_w < 300 or new_h < 300:
+            raise RuntimeError(
+                f"图片压缩至 10MB 以内后尺寸（{new_w}x{new_h}）低于最小限制 300px，无法同时满足两项约束。"
+            )
+        img = img.resize((new_w, new_h), resample=1)
+    else:
+        raise RuntimeError("图片经过 20 次缩放仍超过 10MB，请检查原始图片。")
+
+    buf.seek(0)
+    return base64.b64encode(buf.read()).decode("utf-8")
+
+
+# ── 节点 ──────────────────────────────────────────────────────────────────────
+
+class K3Video:
+    """K3 图生视频 自研"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        required = {
+            "多镜头": (_MULTI_SHOT_OPTIONS, {
+                "default": "禁用",
+                "tooltip": "禁用：单段模式；N个故事板：启用 N 段分镜。",
+            }),
+            "起始帧":    ("IMAGE",),
+            "提示词":    ("STRING", {"multiline": True, "default": ""}),
+            "负向提示词": ("STRING", {"multiline": True, "default": ""}),
+            "时长":      ([5, 10, 15], {"default": 5}),
+            "生成音频":  (["关闭", "打开"], {"default": "关闭"}),
+            "模式":      (_MODES, {"default": "720p"}),
+            "网络线路":  (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+            "seed": ("INT", {
+                "default": 0, "min": 0, "max": 2147483647,
+                "tooltip": "seed 仅控制节点是否重新运行，结果本身不可复现。",
+            }),
+        }
+
+        for i in range(1, 7):
+            required[f"分镜{i}_提示词"] = ("STRING", {
+                "multiline": True, "default": "",
+                "tooltip": f"第 {i} 段分镜提示词，最多 512 字符。",
+            })
+            required[f"分镜{i}_时长"] = ("INT", {
+                "default": 4, "min": 1, "max": 15,
+                "display": "slider",
+                "tooltip": f"第 {i} 段分镜时长（秒）。",
+            })
+
+        return {"required": required}
+
+    RETURN_TYPES  = ("VIDEO",)
+    RETURN_NAMES  = ("视频",)
+    FUNCTION      = "generate"
+    CATEGORY      = "comfyui_o1key/KVideo"
+
+    async def generate(self, 多镜头, 起始帧, 提示词, 负向提示词, 时长, 生成音频, 模式, 网络线路, seed, **kwargs):
+        api_key  = get_api_key_or_raise()
+        base_url = get_base_url_by_route(网络线路)
+        headers  = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type":  "application/json",
+        }
+
+        is_multi  = 多镜头 != "禁用"
+        voice     = "voice" if 生成音频 == "打开" else "novoice"
+        mode_api  = _MODE_MAP[模式]
+        if mode_api == "4k":
+            model_name = f"{_MODEL_BASE}-4k-{时长}s"
+        else:
+            model_name = f"{_MODEL_BASE}-{mode_api}-{时长}s-{voice}"
+
+        # ── 多镜头模式 ────────────────────────────────────────────────
+        if is_multi:
+            shot_count = int(多镜头[0])  # "3个故事板" → 3
+
+            # 收集分镜参数
+            multi_prompt = []
+            for i in range(1, shot_count + 1):
+                p = kwargs.get(f"分镜{i}_提示词", "").strip()
+                d = kwargs.get(f"分镜{i}_时长", 0)
+                if not p:
+                    raise RuntimeError(f"多镜头模式错误：第 {i} 段分镜提示词不能为空。")
+                if d < 1:
+                    raise RuntimeError(f"多镜头模式错误：第 {i} 段分镜时长不能小于 1 秒。")
+                multi_prompt.append({"index": i, "prompt": p, "duration": str(d)})
+
+            # 校验时长总和
+            total = sum(int(s["duration"]) for s in multi_prompt)
+            if total != 时长:
+                raise RuntimeError(
+                    f"多镜头模式错误：各分镜时长之和（{total}s）必须等于总时长（{时长}s）。"
+                )
+
+            metadata: dict = {
+                "multi_shot": "true",
+                "shot_type":  "customize",
+                "multi_prompt": multi_prompt,
+            }
+            if 生成音频 == "打开":
+                metadata["sound"] = "on"
+
+            body: dict = {
+                "model":    model_name,
+                "prompt":   提示词.strip() or " ",
+                "mode":     mode_api,
+                "duration": 时长,
+                "image":    _prepare_image_base64(起始帧),
+                "metadata": metadata,
+            }
+            if 负向提示词.strip():
+                body["negative_prompt"] = 负向提示词.strip()
+
+        # ── 单段图生视频模式 ──────────────────────────────────────────
+        else:
+            if not 提示词.strip():
+                raise RuntimeError("单段模式错误：提示词不能为空。")
+
+            body = {
+                "model":    model_name,
+                "prompt":   提示词.strip(),
+                "mode":     mode_api,
+                "duration": 时长,
+                "image":    _prepare_image_base64(起始帧),
+            }
+            if 负向提示词.strip():
+                body["negative_prompt"] = 负向提示词.strip()
+            if 生成音频 == "打开":
+                body["metadata"] = {"sound": "on"}
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        tag = "多镜头" if is_multi else "图生视频"
+
+        def _stage(s: str):
+            if s == "submitting":
+                print(f"[K3 {tag}] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif s.startswith("submitted:"):
+                print(f"[K3 {tag}] 任务已提交 → {s.split(':', 1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif s == "downloading":
+                print(f"[K3 {tag}] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif s == "done":
+                print(f"[K3 {tag}] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def _progress(pct: int):
+            if pbar: pbar.update_absolute(5 + int(pct * 0.94), 100)
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        tmp_fd, save_path = tempfile.mkstemp(suffix=".mp4", prefix="k3_")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交
+            check_interrupt()
+            _stage("submitting")
+            create_url = f"{base_url}{_ENDPOINT_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url, json=body, headers=headers, prefix="K3 提交: "
+            ))
+            check_interrupt()
+            text = await resp.text()
+            create_resp = json.loads(text)
+
+            task_id = (
+                create_resp.get("task_id")
+                or create_resp.get("id")
+                or create_resp.get("data", {}).get("task_id")
+            )
+            if not task_id:
+                raise RuntimeError(f"API 未返回任务 ID，响应：{create_resp}")
+            _stage(f"submitted:{task_id}")
+
+            # 2. 轮询
+            status_url = f"{base_url}{_ENDPOINT_STATUS.format(task_id=task_id)}"
+            interval   = _POLL_INIT
+            video_url  = None
+
+            while True:
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    text = await resp.text()
+                    if resp.status != 200:
+                        try:
+                            err = json.loads(text)
+                            msg = err.get("error", {}).get("message") or err.get("message") or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                    sr = json.loads(text)
+
+                data   = sr.get("data", sr)
+                status = extract_status(sr)
+
+                pct = extract_progress(sr)
+                print(f"[K3 {tag}] 生成中 {pct}%")
+                _progress(pct)
+
+                if is_success_status(status):
+                    video_url = extract_video_url(sr)
+                    break
+                if is_failure_status(status, sr):
+                    err_msg = extract_error_message(sr)
+                    raise RuntimeError(f"K3 生成失败：{err_msg}")
+
+                await interruptible_sleep(interval)
+                interval = min(interval * 1.5, _POLL_MAX)
+
+            if not video_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{sr}")
+
+            # 3. 下载
+            check_interrupt()
+            _stage("downloading")
+            async with session.get(video_url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                os.close(tmp_fd)
+                with open(save_path, "wb") as f:
+                    async for chunk in resp.content.iter_chunked(8192):
+                        check_interrupt()
+                        f.write(chunk)
+
+        _stage("done")
+
+        if _FOLDER_PATHS_OK:
+            return (InputImpl.VideoFromFile(save_path),)
+        return (save_path,)
+
+
+# ── 节点注册 ──────────────────────────────────────────────────────────────────
+
+NODE_CLASS_MAPPINGS = {
+    "K3Video": K3Video,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "K3Video": "K3 图生视频 自研",
+}
@@ -0,0 +1,283 @@
+"""
+首尾帧 K3 自研节点
+基于 K3 图生视频 自研，去掉分镜功能，新增尾帧可选输入。
+"""
+
+import asyncio
+import json
+import os
+import tempfile
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_async_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+try:
+    from comfy_api.latest import InputImpl
+    import folder_paths
+    _FOLDER_PATHS_OK = True
+except Exception:
+    _FOLDER_PATHS_OK = False
+
+
+# ── 常量 ──────────────────────────────────────────────────────────────────────
+
+_MODEL_BASE = "kling-v3"
+_MODES      = ["720p", "1080p", "4K"]
+_MODE_MAP   = {"720p": "std", "1080p": "pro", "4K": "4k"}
+
+_ENDPOINT_CREATE = "/v1/video/generations"
+_ENDPOINT_STATUS = "/v1/video/generations/{task_id}"
+
+_POLL_INIT = 3
+_POLL_MAX  = 15
+
+
+# ── 工具函数 ───────────────────────────────────────────────────────────────────
+
+def _prepare_image_base64(tensor) -> str:
+    """转换并校验图片，不符合约束时自动等比缩放后返回 base64。"""
+    import io
+    import base64
+
+    pil_list = tensor_to_pil(tensor)
+    img = pil_list[0].convert("RGB")
+    w, h = img.size
+
+    # 1. 宽高比校验
+    ratio = w / h
+    if ratio < 1 / 2.5 or ratio > 2.5:
+        raise RuntimeError(
+            f"图片宽高比 {w}:{h}（{ratio:.2f}）超出允许范围 1:2.5 ~ 2.5:1，请裁剪后重试。"
+        )
+
+    # 2. 最小尺寸：任意边 < 300px 时等比放大
+    if w < 300 or h < 300:
+        scale = max(300 / w, 300 / h)
+        img = img.resize((int(w * scale), int(h * scale)), resample=1)
+
+    # 3. 文件大小：循环等比缩小直到 ≤ 10MB
+    MAX_BYTES = 10 * 1024 * 1024
+    for _ in range(20):
+        buf = io.BytesIO()
+        img.save(buf, format="PNG")
+        if buf.tell() <= MAX_BYTES:
+            break
+        scale = (MAX_BYTES / buf.tell()) ** 0.5 * 0.95
+        new_w = int(img.width * scale)
+        new_h = int(img.height * scale)
+        if new_w < 300 or new_h < 300:
+            raise RuntimeError(
+                f"图片压缩至 10MB 以内后尺寸（{new_w}x{new_h}）低于最小限制 300px，无法同时满足两项约束。"
+            )
+        img = img.resize((new_w, new_h), resample=1)
+    else:
+        raise RuntimeError("图片经过 20 次缩放仍超过 10MB，请检查原始图片。")
+
+    buf.seek(0)
+    return base64.b64encode(buf.read()).decode("utf-8")
+
+
+# ── 节点 ──────────────────────────────────────────────────────────────────────
+
+class K3VideoFirstLast:
+    """首尾帧 K3 自研"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "起始帧":    ("IMAGE",),
+                "提示词":    ("STRING", {"multiline": True, "default": ""}),
+                "负向提示词": ("STRING", {"multiline": True, "default": ""}),
+                "时长":      ([5, 10, 15], {"default": 5}),
+                "生成音频":  (["关闭", "打开"], {"default": "关闭"}),
+                "模式":      (_MODES, {"default": "720p"}),
+                "网络线路":  (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "seed": ("INT", {
+                    "default": 0, "min": 0, "max": 2147483647,
+                    "tooltip": "seed 仅控制节点是否重新运行，结果本身不可复现。",
+                }),
+            },
+            "optional": {
+                "尾帧": ("IMAGE", {"tooltip": "可选。传入后将作为视频尾帧参考。"}),
+            },
+        }
+
+    RETURN_TYPES  = ("VIDEO",)
+    RETURN_NAMES  = ("视频",)
+    FUNCTION      = "generate"
+    CATEGORY      = "comfyui_o1key/KVideo"
+
+    async def generate(self, 起始帧, 提示词, 负向提示词, 时长, 生成音频, 模式, 网络线路, seed, 尾帧=None):
+        api_key  = get_api_key_or_raise()
+        base_url = get_base_url_by_route(网络线路)
+        headers  = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type":  "application/json",
+        }
+
+        voice      = "voice" if 生成音频 == "打开" else "novoice"
+        mode_api   = _MODE_MAP[模式]
+        if mode_api == "4k":
+            model_name = f"{_MODEL_BASE}-4k-{时长}s"
+        else:
+            model_name = f"{_MODEL_BASE}-{mode_api}-{时长}s-{voice}"
+
+        if not 提示词.strip():
+            raise RuntimeError("提示词不能为空。")
+
+        # ── 构建请求体 ────────────────────────────────────────────────
+        body: dict = {
+            "model":    model_name,
+            "prompt":   提示词.strip(),
+            "mode":     mode_api,
+            "duration": 时长,
+            "image":    _prepare_image_base64(起始帧),
+        }
+
+        if 负向提示词.strip():
+            body["negative_prompt"] = 负向提示词.strip()
+
+        # metadata：尾帧 + 音频
+        metadata: dict = {}
+        if 尾帧 is not None:
+            metadata["image_tail"] = _prepare_image_base64(尾帧)
+        if 生成音频 == "打开":
+            metadata["sound"] = "on"
+        if metadata:
+            body["metadata"] = metadata
+
+        # generate_audio 字段（非 metadata 路径）
+        if 生成音频 == "打开" and not metadata.get("sound"):
+            body["generate_audio"] = True
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def _stage(s: str):
+            if s == "submitting":
+                print("[K3 首尾帧] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif s.startswith("submitted:"):
+                print(f"[K3 首尾帧] 任务已提交 → {s.split(':', 1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif s == "downloading":
+                print("[K3 首尾帧] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif s == "done":
+                print("[K3 首尾帧] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def _progress(pct: int):
+            if pbar: pbar.update_absolute(5 + int(pct * 0.94), 100)
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        tmp_fd, save_path = tempfile.mkstemp(suffix=".mp4", prefix="k3fl_")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交
+            check_interrupt()
+            _stage("submitting")
+            create_url = f"{base_url}{_ENDPOINT_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url, json=body, headers=headers, prefix="K3 首尾帧提交: "
+            ))
+            check_interrupt()
+            text = await resp.text()
+            create_resp = json.loads(text)
+
+            task_id = (
+                create_resp.get("task_id")
+                or create_resp.get("id")
+                or create_resp.get("data", {}).get("task_id")
+            )
+            if not task_id:
+                raise RuntimeError(f"API 未返回任务 ID，响应：{create_resp}")
+            _stage(f"submitted:{task_id}")
+
+            # 2. 轮询
+            status_url = f"{base_url}{_ENDPOINT_STATUS.format(task_id=task_id)}"
+            interval   = _POLL_INIT
+            video_url  = None
+
+            while True:
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    text = await resp.text()
+                    if resp.status != 200:
+                        try:
+                            err = json.loads(text)
+                            msg = err.get("error", {}).get("message") or err.get("message") or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                    sr = json.loads(text)
+
+                data   = sr.get("data", sr)
+                status = extract_status(sr)
+
+                pct = extract_progress(sr)
+                print(f"[K3 首尾帧] 生成中 {pct}%")
+                _progress(pct)
+
+                if is_success_status(status):
+                    video_url = extract_video_url(sr)
+                    break
+                if is_failure_status(status, sr):
+                    err_msg = extract_error_message(sr)
+                    raise RuntimeError(f"K3 首尾帧生成失败：{err_msg}")
+
+                await interruptible_sleep(interval)
+                interval = min(interval * 1.5, _POLL_MAX)
+
+            if not video_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{sr}")
+
+            # 3. 下载
+            check_interrupt()
+            _stage("downloading")
+            async with session.get(video_url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                os.close(tmp_fd)
+                with open(save_path, "wb") as f:
+                    async for chunk in resp.content.iter_chunked(8192):
+                        check_interrupt()
+                        f.write(chunk)
+
+        _stage("done")
+
+        if _FOLDER_PATHS_OK:
+            return (InputImpl.VideoFromFile(save_path),)
+        return (save_path,)
+
+
+# ── 节点注册 ──────────────────────────────────────────────────────────────────
+
+NODE_CLASS_MAPPINGS = {
+    "K3VideoFirstLast": K3VideoFirstLast,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "K3VideoFirstLast": "首尾帧 K3 自研",
+}
@@ -0,0 +1,254 @@
+"""
+K26 图生视频节点
+"""
+
+import asyncio
+import json
+import math
+import os
+import tempfile
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+try:
+    from comfy_api.latest import InputImpl
+    import folder_paths
+    _FOLDER_PATHS_OK = True
+except ImportError:
+    _FOLDER_PATHS_OK = False
+
+# 模型基础名，运行时动态拼接完整名称
+_MODEL_BASE = "kling-v2-6"
+
+# API 端点
+_ENDPOINT_CREATE = "/v1/video/generations"
+_ENDPOINT_STATUS = "/v1/video/generations/{task_id}"
+
+_POLL_INIT     = 3
+_POLL_MAX      = 15
+
+
+def _image_to_base64(tensor, scale=1.0) -> str:
+    from PIL import Image
+    pil = tensor_to_pil(tensor)
+    img = pil[0]
+    if scale < 1.0:
+        w, h = img.size
+        new_w = max(1, int(w * scale))
+        new_h = max(1, int(h * scale))
+        img = img.resize((new_w, new_h), Image.LANCZOS)
+    return encode_image_to_base64(img, format="PNG")
+
+
+class KVideoFirstLast:
+    """K26 图生视频节点（首尾帧）"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "起始帧":   ("IMAGE",),
+                "提示词":   ("STRING", {"multiline": True, "default": ""}),
+                "模式":     (["1080p"],),
+                "时长":     ([5, 10],),
+                "生成音频": (["关闭", "打开"], {"default": "关闭"}),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "seed": ("INT", {
+                    "default": 0, "min": 0, "max": 2147483647,
+                    "tooltip": "seed 仅控制节点是否重新运行，结果本身不可复现。",
+                }),
+            },
+            "optional": {
+                "尾帧":     ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/KVideo"
+
+    async def generate(self, 起始帧, 提示词, 模式, 时长, 生成音频="关闭", 网络线路="全球加速", 尾帧=None, seed=0):
+        api_key  = get_api_key_or_raise()
+        base_url = get_base_url_by_route(网络线路)
+        headers  = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type":  "application/json",
+        }
+
+        # ── 动态拼接模型名 ────────────────────────────────────────────
+        mode_api   = "pro"  # 1080p 映射为 pro
+        voice      = "voice" if 生成音频 == "打开" else "novoice"
+        model_name = f"{_MODEL_BASE}-{mode_api}-{时长}s-{voice}"
+
+        # ── 构建请求体（超过 10MB 自动缩放图片）────────────────────────
+        MAX_BODY  = 10 * 1024 * 1024
+        scale     = 1.0
+
+        print(f"[K26 图生视频] 请求体大小限制: 10MB，超出将自动缩放图片")
+
+        while True:
+            body = {
+                "model":    model_name,
+                "prompt":   提示词.strip(),
+                "image":    _image_to_base64(起始帧, scale),
+                "mode":     mode_api,
+                "duration": 时长,
+            }
+            metadata = {}
+            if 尾帧 is not None:
+                metadata["image_tail"] = _image_to_base64(尾帧, scale)
+            if 生成音频 == "打开":
+                metadata["sound"] = "on"
+            if metadata:
+                body["metadata"] = metadata
+
+            body_str  = json.dumps(body, ensure_ascii=False)
+            body_size = len(body_str.encode("utf-8"))
+
+            if body_size <= MAX_BODY:
+                print(f"[K26 图生视频] 请求体大小: {body_size / 1024 / 1024:.2f}MB"
+                      + (f"（已缩放至 {scale:.1%}）" if scale < 1.0 else ""))
+                break
+
+            # 等比缩放：图片像素面积与 base64 长度近似线性
+            target_ratio = MAX_BODY / body_size
+            scale = scale * math.sqrt(target_ratio) * 0.95  # 5% 安全余量
+
+            if scale < 0.01:
+                raise RuntimeError("图片缩放后仍超过10MB限制，请使用更小的参考图")
+
+            w, h = tensor_to_pil(起始帧)[0].size
+            print(f"[K26 图生视频] 请求体 {body_size / 1024 / 1024:.2f}MB 超限，"
+                  f"自动缩放至 {scale:.1%}（{int(w * scale)}x{int(h * scale)}）")
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def _stage(s: str):
+            if s == "submitting":
+                print("[K26 图生视频] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif s.startswith("submitted:"):
+                print(f"[K26 图生视频] 任务已提交 → {s.split(':', 1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif s == "downloading":
+                print("[K26 图生视频] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif s == "done":
+                print("[K26 图生视频] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def _progress(pct: int):
+            if pbar: pbar.update_absolute(5 + int(pct * 0.94), 100)
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        tmp_fd, save_path = tempfile.mkstemp(suffix=".mp4", prefix="k26_")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交
+            check_interrupt()
+            _stage("submitting")
+            create_url = f"{base_url}{_ENDPOINT_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url, json=body, headers=headers, prefix="K26 提交: "
+            ))
+            check_interrupt()
+            text = await resp.text()
+            create_resp = json.loads(text)
+
+            task_id = (
+                create_resp.get("task_id")
+                or create_resp.get("id")
+                or create_resp.get("data", {}).get("task_id")
+            )
+            if not task_id:
+                raise RuntimeError(f"API 未返回任务 ID，响应：{create_resp}")
+            _stage(f"submitted:{task_id}")
+
+            # 2. 轮询
+            status_url = f"{base_url}{_ENDPOINT_STATUS.format(task_id=task_id)}"
+            interval   = _POLL_INIT
+            video_url  = None
+
+            while True:
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    text = await resp.text()
+                    if resp.status != 200:
+                        try:
+                            err = json.loads(text)
+                            msg = err.get("error", {}).get("message") or err.get("message") or text
+                        except Exception:
+                            msg = text
+                        raise RuntimeError(f"状态查询失败 ({resp.status}): {msg}")
+                    sr = json.loads(text)
+
+                data   = sr.get("data", sr)
+                status = extract_status(sr)
+
+                pct = extract_progress(sr)
+                print(f"[K26 图生视频] 生成中 {pct}%")
+                _progress(pct)
+
+                if is_success_status(status):
+                    # 提取视频 URL
+                    video_url = extract_video_url(sr)
+                    break
+                if is_failure_status(status, sr):
+                    err_msg = extract_error_message(sr)
+                    raise RuntimeError(f"K26 生成失败：{err_msg}")
+
+                await interruptible_sleep(interval)
+                interval = min(interval * 1.5, _POLL_MAX)
+
+            if not video_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{sr}")
+
+            # 3. 下载
+            check_interrupt()
+            _stage("downloading")
+            async with session.get(video_url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                os.close(tmp_fd)
+                with open(save_path, "wb") as f:
+                    async for chunk in resp.content.iter_chunked(8192):
+                        check_interrupt()
+                        f.write(chunk)
+
+        _stage("done")
+
+        if _FOLDER_PATHS_OK:
+            return (InputImpl.VideoFromFile(save_path),)
+        return (save_path,)
+
+
+NODE_CLASS_MAPPINGS = {
+    "KVideoFirstLast": KVideoFirstLast,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "KVideoFirstLast": "K26 图生视频（首尾帧）",
+}
@@ -0,0 +1,242 @@
+"""
+K26 图生视频节点
+支持 720p 和 1080p 模式
+"""
+
+import asyncio
+import json
+import math
+import os
+import tempfile
+
+import aiohttp
+
+from ..utils.config import get_api_key_or_raise, get_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.http_error import async_request_with_retry
+from ..utils.video_task import (
+    check_interrupt,
+    extract_error_message,
+    extract_progress,
+    extract_status,
+    extract_video_url,
+    interruptible_sleep,
+    is_failure_status,
+    is_success_status,
+    run_with_interrupt,
+)
+
+try:
+    from comfy_api.latest import InputImpl
+    import folder_paths
+    _FOLDER_PATHS_OK = True
+except ImportError:
+    _FOLDER_PATHS_OK = False
+
+# 模型基础名，运行时动态拼接完整名称
+_MODEL_BASE = "kling-v2-6"
+
+# API 端点
+_ENDPOINT_CREATE = "/v1/video/generations"
+_ENDPOINT_STATUS = "/v1/video/generations/{task_id}"
+
+_POLL_INIT     = 3
+_POLL_MAX      = 15
+
+
+def _image_to_base64(tensor, scale=1.0) -> str:
+    from PIL import Image
+    pil = tensor_to_pil(tensor)
+    img = pil[0]
+    if scale < 1.0:
+        w, h = img.size
+        new_w = max(1, int(w * scale))
+        new_h = max(1, int(h * scale))
+        img = img.resize((new_w, new_h), Image.LANCZOS)
+    return encode_image_to_base64(img, format="PNG")
+
+
+class KVideoImage2Video:
+    """K26 图生视频节点"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "起始帧":   ("IMAGE",),
+                "提示词":   ("STRING", {"multiline": True, "default": ""}),
+                "模式":     (["720p", "1080p"], {"default": "720p"}),
+                "时长":     ([5, 10], {"default": 5}),
+                "生成音频": (["关闭", "打开"], {"default": "关闭"}),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "seed": ("INT", {
+                    "default": 0, "min": 0, "max": 2147483647,
+                    "tooltip": "seed 仅控制节点是否重新运行，结果本身不可复现。",
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/KVideo"
+
+    async def generate(self, 起始帧, 提示词, 模式, 时长, 生成音频="关闭", 网络线路="全球加速", seed=0):
+        if 模式 == "720p" and 生成音频 == "打开":
+            raise RuntimeError("K26 仅1080p支持音频，请将模式切换为1080p或关闭生成音频。")
+
+        api_key  = get_api_key_or_raise()
+        base_url = get_base_url_by_route(网络线路)
+        headers  = {
+            "Authorization": f"Bearer {api_key}",
+            "Content-Type":  "application/json",
+        }
+
+        # ── 动态拼接模型名 ────────────────────────────────────────────
+        mode_api   = "std" if 模式 == "720p" else "pro"
+        voice      = "voice" if 生成音频 == "打开" else "novoice"
+        model_name = f"{_MODEL_BASE}-{mode_api}-{时长}s-{voice}"
+
+        # ── 构建请求体（超过 10MB 自动缩放图片）────────────────────────
+        MAX_BODY  = 10 * 1024 * 1024
+        scale     = 1.0
+
+        print(f"[K26 图生视频] 请求体大小限制: 10MB，超出将自动缩放图片")
+
+        while True:
+            body = {
+                "model":    model_name,
+                "prompt":   提示词.strip(),
+                "image":    _image_to_base64(起始帧, scale),
+                "mode":     mode_api,
+                "duration": 时长,
+            }
+            if 生成音频 == "打开":
+                body["metadata"] = {"sound": "on"}
+
+            body_str  = json.dumps(body, ensure_ascii=False)
+            body_size = len(body_str.encode("utf-8"))
+
+            if body_size <= MAX_BODY:
+                print(f"[K26 图生视频] 请求体大小: {body_size / 1024 / 1024:.2f}MB"
+                      + (f"（已缩放至 {scale:.1%}）" if scale < 1.0 else ""))
+                break
+
+            # 等比缩放：图片像素面积与 base64 长度近似线性
+            target_ratio = MAX_BODY / body_size
+            scale = scale * math.sqrt(target_ratio) * 0.95  # 5% 安全余量
+
+            if scale < 0.01:
+                raise RuntimeError("图片缩放后仍超过10MB限制，请使用更小的参考图")
+
+            w, h = tensor_to_pil(起始帧)[0].size
+            print(f"[K26 图生视频] 请求体 {body_size / 1024 / 1024:.2f}MB 超限，"
+                  f"自动缩放至 {scale:.1%}（{int(w * scale)}x{int(h * scale)}）")
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def _stage(s: str):
+            if s == "submitting":
+                print("[K26 图生视频] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif s.startswith("submitted:"):
+                print(f"[K26 图生视频] 任务已提交 → {s.split(':', 1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif s == "downloading":
+                print("[K26 图生视频] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif s == "done":
+                print("[K26 图生视频] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def _progress(pct: int):
+            if pbar: pbar.update_absolute(5 + int(pct * 0.94), 100)
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        tmp_fd, save_path = tempfile.mkstemp(suffix=".mp4", prefix="k26_")
+
+        connector = aiohttp.TCPConnector(ssl=False, force_close=True)
+        async with aiohttp.ClientSession(connector=connector) as session:
+
+            # 1. 提交
+            check_interrupt()
+            _stage("submitting")
+            create_url = f"{base_url}{_ENDPOINT_CREATE}"
+            resp = await run_with_interrupt(async_request_with_retry(
+                session, "POST", create_url, json=body, headers=headers, prefix="K26 图生视频提交: "
+            ))
+            check_interrupt()
+            sr = await resp.json()
+
+            task_id = sr.get("task_id") or sr.get("id")
+            if not task_id:
+                raise RuntimeError(f"API 未返回 task_id，响应：{sr}")
+
+            _stage(f"submitted:{task_id}")
+
+            # 2. 轮询
+            status_url = f"{base_url}{_ENDPOINT_STATUS.format(task_id=task_id)}"
+            interval   = _POLL_INIT
+            video_url  = None
+
+            while True:
+                await interruptible_sleep(interval)
+
+                check_interrupt()
+                async with session.get(status_url, headers=headers) as resp:
+                    if resp.status != 200:
+                        err_text = await resp.text()
+                        raise RuntimeError(f"查询失败 ({resp.status}): {err_text}")
+                    sr = await resp.json()
+
+                data   = sr.get("data", {}) or {}
+                status = extract_status(sr)
+
+                pct = extract_progress(sr)
+                print(f"[K26 图生视频] 生成中 {pct}%")
+                _progress(pct)
+
+                if is_success_status(status):
+                    # 提取视频 URL
+                    video_url = extract_video_url(sr)
+                    break
+                if is_failure_status(status, sr):
+                    err_msg = extract_error_message(sr)
+                    raise RuntimeError(f"K26 生成失败：{err_msg}")
+
+                interval = min(interval * 1.5, _POLL_MAX)
+
+            if not video_url:
+                raise RuntimeError(f"API 未返回视频 URL，响应：{sr}")
+
+            # 3. 下载
+            check_interrupt()
+            _stage("downloading")
+            async with session.get(video_url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    raise RuntimeError(f"视频下载失败 ({resp.status})")
+                os.close(tmp_fd)
+                with open(save_path, "wb") as f:
+                    async for chunk in resp.content.iter_chunked(8192):
+                        check_interrupt()
+                        f.write(chunk)
+
+        _stage("done")
+
+        if _FOLDER_PATHS_OK:
+            return (InputImpl.VideoFromFile(save_path),)
+        return (save_path,)
+
+
+NODE_CLASS_MAPPINGS = {
+    "KVideoImage2Video": KVideoImage2Video,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "KVideoImage2Video": "K26 图生视频",
+}
@@ -3,8 +3,36 @@
 包含所有 ComfyUI 自定义节点的实现
 """

-from .nano_banana_pro import NanoBananaPro
-from .batch_nano_banana_pro import BatchNanoBananaPro
+from .stream_preview import StreamPreview
+from .nano_banana import NanoBanana
+NanoBananaPro = NanoBanana
+from .batch_nano_banana import BatchNanoBananaPro
 from .google_gemini import GoogleGemini
+from .load_file import LoadFile
+from .image_stitch_pro import ImageStitchPro
+from .remove_metadata import BatchCleanMetadata
+from .video_preview import VideoPreview
+from .kling_video import KlingVideo, KlingFirstLastFrame, KlingMotionControlTest, AspectRatioPreset
+from .veo_video import GoogleVeo
+from .newapi_veo_video import Google31Video
+from .flux_edit import FluxImageEdit
+from .universal_llm import UniversalLLMChat
+from .batch_images_o1key import BatchImagesO1key
+from .seedance_video import Seedance, SeedanceMultiModal
+from .nano_banana_v2 import NanoBananaV2, NanoBananaV2Batch, AsyncImageGenerator, BatchAsyncImageGenerator
+from .doubao_image import DoubaoImage
+from .gpt_image import O1keyGPTImage, O1keyGPTImageBatch
+from .grok_image import O1keyGrokImage
+from .grok_video import O1keyGrokVideo
+from .K_video_firstlast import KVideoFirstLast
+from .K_video_image2video import KVideoImage2Video
+from .K3_video import K3Video
+from .K3_video_firstlast import K3VideoFirstLast
+from .K3_motion_control import K3MotionControl, K3MotionVideoCheck
+from .save_image_format import SaveImageFormat
+from .save_psd import O1keySavePSD
+from .remove_bg import O1keyRemoveBackground
+from .color_remove_bg import O1keyColorRemoveBG
+from .grid_splitter import O1keyGridSplitter

-__all__ = ['NanoBananaPro', 'BatchNanoBananaPro', 'GoogleGemini']
+__all__ = ['NanoBananaV2', 'NanoBananaV2Batch', 'NanoBanana', 'BatchNanoBananaPro', 'GoogleGemini', 'LoadFile', 'ImageStitchPro', 'BatchCleanMetadata', 'VideoPreview', 'KlingVideo', 'KlingFirstLastFrame', 'KlingMotionControlTest', 'AspectRatioPreset', 'GoogleVeo', 'Google31Video', 'FluxImageEdit', 'UniversalLLMChat', 'BatchImagesO1key', 'Seedance', 'SeedanceMultiModal', 'StreamPreview', 'DoubaoImage', 'O1keyGPTImage', 'O1keyGPTImageBatch', 'O1keyGrokImage', 'O1keyGrokVideo', 'KVideoFirstLast', 'KVideoImage2Video', 'K3Video', 'K3VideoFirstLast', 'K3MotionControl', 'K3MotionVideoCheck', 'AsyncImageGenerator', 'BatchAsyncImageGenerator', 'SaveImageFormat', 'O1keySavePSD', 'O1keyRemoveBackground', 'O1keyColorRemoveBG', 'O1keyGridSplitter']
@@ -0,0 +1,80 @@
+"""
+批量图像（o1key）节点
+复刻 ComfyUI 原生「批量图像」节点的动态输入行为：
+
+- 默认显示 2 个图像输入端口（图1, 图2）
+- 当最后一个端口连上图像后，自动追加新端口
+- 断开连线后，多余的端口自动消失，最少保留 2 个
+
+与原生节点的区别：
+  原生节点会把所有图像强制 resize 到第一张的分辨率再合并为单一 tensor。
+  本节点保留每张图的原始分辨率，以 list[Tensor] 形式输出（is_output_list）。
+  下游节点（如「多分辨率图像预览」）需开启 INPUT_IS_LIST 才能正确接收。
+
+实现方式：使用 V3 API 的 io.Autogrow.TemplateNames，
+框架原生支持动态 slot 增减，无需编写任何 JS 扩展。
+"""
+
+import torch
+from comfy_api.latest import io
+
+# 预生成 50 个端口名：图1, 图2, ..., 图50
+_SLOT_NAMES = [f"图{i}" for i in range(1, 51)]
+
+
+class BatchImagesO1key(io.ComfyNode):
+    """
+    批量图像（o1key）
+
+    - 动态输入端口（默认 2 个，最多 50 个），端口名为 图1、图2、图3...
+    - 连接最后一个端口时自动增加新端口
+    - 断开后自动减少，保持界面整洁
+    - 保留每张图的原始分辨率，不做任何 resize / 裁剪
+    - 输出为图像列表，可直接接入「多分辨率图像预览」节点
+    """
+
+    @classmethod
+    def define_schema(cls):
+        autogrow_template = io.Autogrow.TemplateNames(
+            input=io.Image.Input("image"),
+            names=_SLOT_NAMES,
+            min=2,
+        )
+        return io.Schema(
+            node_id="BatchImagesO1key",
+            display_name="加载图像（批量）",
+            category="image",
+            description=(
+                "将多个独立图像收集为图像列表输出，保留每张图的原始分辨率。\n"
+                "• 默认显示 2 个输入端口（图1、图2），连接最后一个后自动追加新端口\n"
+                "• 断开连线后端口自动减少，最少保留 2 个\n"
+                "• 不做任何 resize / 裁剪，原图尺寸原样输出\n"
+                "• 输出为图像列表，可直接接入「多分辨率图像预览」节点"
+            ),
+            search_aliases=["批量图像", "batch images", "合并图像", "图像合并", "stack images"],
+            inputs=[
+                io.Autogrow.Input("images", template=autogrow_template)
+            ],
+            outputs=[
+                io.Image.Output(display_name="图像", is_output_list=True),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, images: io.Autogrow.Type) -> io.NodeOutput:
+        # images 是 dict，key 为 "图1", "图2", ... ；未连接的 slot 值为 None
+        tensors = [v for v in images.values() if v is not None]
+
+        if not tensors:
+            raise ValueError("批量图像（o1key）：请至少连接一张图像")
+
+        for i, t in enumerate(tensors):
+            h, w = t.shape[1], t.shape[2]
+            print(f"批量图像（o1key）：图{i + 1} → {w}×{h}，shape={list(t.shape)}")
+
+        print(f"批量图像（o1key）：共收集 {len(tensors)} 张，原始分辨率原样输出")
+
+        # 以 list[Tensor] 形式返回，每张图保持自身分辨率
+        return io.NodeOutput(tensors)
+
+
@@ -1,751 +0,0 @@
-"""
-批量 Nano Banana Pro 节点
-ComfyUI 自定义节点，用于批量处理图像生成任务
-支持多文件夹加载、1:1/笛卡尔积配对、智能命名保存
-"""
-
-import time
-import math
-import random
-import asyncio
-import aiohttp
-from concurrent.futures import ThreadPoolExecutor
-from typing import Optional, Tuple, List
-from PIL import Image
-
-import torch
-import numpy as np
-
-from ..utils.image_utils import tensor_to_pil, pil_to_tensor
-from ..utils.file_utils import (
-    ImageInfo,
-    load_images_from_folder,
-    pair_images_indexed,
-    pair_images_cartesian,
-    generate_output_filename,
-    save_image
-)
-from ..clients.gemini_client import GeminiAPIClient
-from ..models_config import get_enabled_models
-
-# 导入 ComfyUI 原生进度条
-try:
-    from comfy.utils import ProgressBar
-    PROGRESS_BAR_AVAILABLE = True
-except ImportError:
-    PROGRESS_BAR_AVAILABLE = False
-    print("⚠️ BatchNanoBananaPro: comfy.utils.ProgressBar 不可用，将只使用终端进度显示")
-
-
-class BatchNanoBananaPro:
-    """
-    批量 Nano Banana Pro 节点
-    
-    功能：
-    - 从多个文件夹加载图片
-    - 支持三种配对模式：
-      * 1:1 - 索引配对（文件夹之间按位置配对）
-      * 1*N - 笛卡尔积配对（所有可能组合）
-      * 不配对 - 固定参考图模式（文件夹图片依次与所有参考图组合）
-    - 批量调用 API 生成图像
-    - 智能命名保存（保留原始文件名）
-    - 并发控制（默认最大 100）
-    
-    注意：
-    - 「不配对」模式只支持单个文件夹
-    - 支持的模型列表从 models_config.py 动态加载
-    - 要添加/禁用模型，请编辑 models_config.py 文件
-    """
-    
-    # 支持的模型列表（从配置文件动态加载）
-    MODELS = None  # 将在 INPUT_TYPES 中动态获取
-    
-    # 支持的宽高比列表
-    ASPECT_RATIOS = [
-        "1:1", "4:3", "3:4", "16:9", "9:16",
-        "2:3", "3:2", "4:5", "5:4", "21:9"
-    ]
-    
-    # 支持的分辨率列表
-    RESOLUTIONS = ["1K", "2K", "4K"]
-    
-    # 配对模式
-    PAIRING_MODES = ["1:1", "1*N", "不配对"]
-    
-    def __init__(self):
-        """初始化节点"""
-        self.client = None
-    
-    def resize_to_megapixels(
-        self,
-        image: Image.Image,
-        target_megapixels: float
-    ) -> Image.Image:
-        """
-        将图像缩放到指定的总像素数，保持纵横比
-        
-        Args:
-            image: PIL Image 对象
-            target_megapixels: 目标像素数（百万像素）
-        
-        Returns:
-            缩放后的 PIL Image
-        
-        Example:
-            >>> resized = self.resize_to_megapixels(img, 2.0)  # 缩放到2百万像素
-        """
-        # 计算当前像素数
-        current_pixels = image.width * image.height
-        target_pixels = int(target_megapixels * 1_000_000)
-        
-        # 如果当前像素数已经接近目标，则不缩放
-        if abs(current_pixels - target_pixels) / target_pixels < 0.05:
-            return image
-        
-        # 计算缩放比例
-        scale = (target_pixels / current_pixels) ** 0.5
-        
-        # 计算新尺寸
-        new_width = int(image.width * scale)
-        new_height = int(image.height * scale)
-        
-        # 确保至少为1像素
-        new_width = max(1, new_width)
-        new_height = max(1, new_height)
-        
-        # 使用 Lanczos 重采样
-        resized_image = image.resize((new_width, new_height), Image.Resampling.LANCZOS)
-        
-        return resized_image
-    
-    @classmethod
-    def INPUT_TYPES(cls):
-        """
-        定义输入参数
-        
-        ComfyUI 节点规范：
-        - required: 必选参数
-        - optional: 可选参数
-        """
-        # 从配置文件动态获取启用的模型列表
-        enabled_models = get_enabled_models()
-        
-        # 如果没有启用的模型，使用空列表（会导致节点不可用，提示用户配置）
-        if not enabled_models:
-            enabled_models = ["请在 models_config.py 中启用至少一个模型"]
-        
-        # 创建9个独立的图像输入
-        optional_inputs = {}
-        for i in range(1, 10):  # 1-9
-            optional_inputs[f"参考图{i}"] = ("IMAGE",)
-        
-        return {
-            "required": {
-                "prompt": ("STRING", {
-                    "default": "一个中国女子的OOTD",
-                    "multiline": True
-                }),
-                "模型": (enabled_models, {
-                    "default": enabled_models[0]
-                }),
-                "宽高比": (cls.ASPECT_RATIOS, {
-                    "default": "1:1"
-                }),
-                "分辨率": (cls.RESOLUTIONS, {
-                    "default": "2K"
-                }),
-                "像素缩放": ("BOOLEAN", {
-                    "default": False
-                }),
-                "分辨率像素": ("FLOAT", {
-                    "default": 1.0,
-                    "min": 0.1,
-                    "max": 100.0,
-                    "step": 0.1,
-                    "display": "number"
-                }),
-                "seed": ("INT", {
-                    "default": 0,
-                    "min": 0,
-                    "max": 0xffffffffffffffff
-                }),
-                "文件夹1": ("STRING", {
-                    "default": "",
-                    "multiline": False
-                }),
-                "文件夹2": ("STRING", {
-                    "default": "",
-                    "multiline": False
-                }),
-                "文件夹3": ("STRING", {
-                    "default": "",
-                    "multiline": False
-                }),
-                "文件夹4": ("STRING", {
-                    "default": "",
-                    "multiline": False
-                }),
-                "保存路径": ("STRING", {
-                    "default": "",
-                    "multiline": False
-                }),
-                "图片配对模式": (cls.PAIRING_MODES, {
-                    "default": "不配对"
-                })
-            },
-            "optional": optional_inputs
-        }
-    
-    # 返回值类型
-    RETURN_TYPES = ("IMAGE",)
-    RETURN_NAMES = ("输出图像",)
-    
-    # 执行函数名
-    FUNCTION = "process_batch"
-    
-    # 节点分类
-    CATEGORY = "image/batch"
-    
-    def _load_folders(
-        self,
-        folder1: str,
-        folder2: Optional[str],
-        folder3: Optional[str],
-        folder4: Optional[str],
-        enable_scaling: bool,
-        target_megapixels: float
-    ) -> List[List[ImageInfo]]:
-        """
-        加载所有文件夹中的图片
-        
-        Args:
-            folder1-4: 文件夹路径
-            enable_scaling: 是否启用像素缩放
-            target_megapixels: 目标像素数（百万像素）
-        
-        Returns:
-            图片列表的列表
-        """
-        folders = [folder1, folder2, folder3, folder4]
-        all_images = []
-        
-        for i, folder in enumerate(folders, 1):
-            if folder and folder.strip():
-                try:
-                    images = load_images_from_folder(folder)
-                    if images:
-                        # 应用像素缩放
-                        if enable_scaling:
-                            scaled_images = []
-                            for img_info in images:
-                                scaled_img = self.resize_to_megapixels(
-                                    img_info.image,
-                                    target_megapixels
-                                )
-                                # 创建新的 ImageInfo，保留其他元数据
-                                scaled_info = ImageInfo(
-                                    image=scaled_img,
-                                    filename=img_info.filename,
-                                    extension=img_info.extension,
-                                    source_path=img_info.source_path
-                                )
-                                scaled_images.append(scaled_info)
-                            images = scaled_images
-                        
-                        all_images.append(images)
-                        print(f"BatchNanoBananaPro: 文件夹{i} 加载了 {len(images)} 张图片")
-                    else:
-                        print(f"BatchNanoBananaPro: 文件夹{i} 为空或没有有效图片")
-                except ValueError as e:
-                    print(f"BatchNanoBananaPro: 文件夹{i} 加载失败 - {e}")
-        
-        return all_images
-    
-    def _create_pairs(
-        self,
-        image_lists: List[List[ImageInfo]],
-        pairing_mode: str,
-        manual_images: Optional[List[ImageInfo]] = None
-    ) -> List[Tuple[ImageInfo, ...]]:
-        """
-        根据配对模式创建图片组合
-        
-        Args:
-            image_lists: 从文件夹加载的图片列表
-            pairing_mode: 配对模式 (1:1, 1*N, 不配对)
-            manual_images: 手动输入的参考图
-        
-        Returns:
-            配对后的元组列表
-        
-        Raises:
-            ValueError: 不配对模式下填入多个文件夹时
-        """
-        # === 新模式：不配对 ===
-        if pairing_mode == "不配对":
-            # 验证：只支持单个文件夹
-            if len(image_lists) > 1:
-                raise ValueError("「不配对」模式只支持单个文件夹，请清空其他文件夹路径")
-            
-            # 场景1：有文件夹 + 有参考图
-            if image_lists and manual_images:
-                folder_images = image_lists[0]
-                # 每张文件夹图片 + 所有参考图
-                pairs = []
-                for img in folder_images:
-                    pair = (img,) + tuple(manual_images)
-                    pairs.append(pair)
-                return pairs
-            
-            # 场景2：有文件夹 + 无参考图
-            elif image_lists:
-                # 每张图片单独成组
-                return [(img,) for img in image_lists[0]]
-            
-            # 场景3：无文件夹 + 有参考图
-            elif manual_images:
-                # 每张参考图单独成组
-                return [(img,) for img in manual_images]
-            
-            else:
-                return []
-        
-        # === 原有逻辑：1:1 和 1*N ===
-        # 如果有手动参考图，添加到列表中（所有参考图作为一个列表）
-        if manual_images:
-            image_lists.append(manual_images)
-        
-        if not image_lists:
-            return []
-        
-        # 如果只有一个列表，直接返回每个图片作为单元素元组
-        if len(image_lists) == 1:
-            return [(img,) for img in image_lists[0]]
-        
-        # 根据配对模式选择配对函数
-        if pairing_mode == "1:1":
-            pairs = pair_images_indexed(*image_lists)
-        else:  # 1*N
-            pairs = pair_images_cartesian(*image_lists)
-        
-        return pairs
-    
-    async def _generate_single_task(
-        self,
-        client: GeminiAPIClient,
-        session: aiohttp.ClientSession,
-        prompt: str,
-        model: str,
-        resolution: str,
-        aspect_ratio: str,
-        images: List[ImageInfo],
-        output_folder: str,
-        task_index: int
-    ) -> dict:
-        """
-        执行单个生成任务
-        
-        Args:
-            client: API 客户端
-            session: aiohttp 会话
-            prompt: 提示词
-            model: 模型名称
-            resolution: 分辨率
-            aspect_ratio: 宽高比
-            images: 输入图片列表
-            output_folder: 输出文件夹
-            task_index: 任务索引
-        
-        Returns:
-            包含结果信息的字典
-        """
-        result = {
-            "task_index": task_index,
-            "success": False,
-            "generated_count": 0,
-            "saved_files": [],
-            "error": None
-        }
-        
-        try:
-            # 准备输入图片
-            input_pil_images = [info.image for info in images]
-            
-            # 调用 API 生成图片（固定生成1次）
-            generated_images = []
-            try:
-                gen_result = await client.generate_single_async(
-                    prompt=prompt,
-                    model=model,
-                    resolution=resolution,
-                    aspect_ratio=aspect_ratio,
-                    images=input_pil_images,
-                    session=session
-                )
-                if gen_result:
-                    generated_images.extend(gen_result)
-            except Exception as e:
-                error_msg = str(e)
-                print(f"BatchNanoBananaPro: 任务 {task_index + 1} 生成失败 - {error_msg}")
-                result["error"] = error_msg
-            
-            # 保存生成的图片
-            for i, gen_img in enumerate(generated_images):
-                # 使用任务索引作为唯一标识，确保并发安全
-                output_path = generate_output_filename(
-                    source_images=list(images),
-                    batch_index=i,
-                    output_folder=output_folder,
-                    extension=".png",
-                    task_id=f"task{task_index}"
-                )
-                save_image(gen_img, output_path)
-                result["saved_files"].append(output_path)
-            
-            # 只有生成了图片才标记为成功
-            if len(generated_images) > 0:
-                result["success"] = True
-            result["generated_count"] = len(generated_images)
-            
-        except Exception as e:
-            result["error"] = str(e)
-        
-        return result
-    
-    async def _process_batch_async(
-        self,
-        pairs: List[Tuple[ImageInfo, ...]],
-        prompt: str,
-        model: str,
-        resolution: str,
-        aspect_ratio: str,
-        output_folder: str,
-        pbar=None
-    ) -> List[dict]:
-        """
-        异步批量处理所有任务
-        
-        Args:
-            pairs: 配对后的图片组合
-            prompt: 提示词
-            model: 模型名称
-            resolution: 分辨率
-            aspect_ratio: 宽高比
-            output_folder: 输出文件夹
-        
-        Returns:
-            所有任务的结果列表
-        """
-        if self.client is None:
-            self.client = GeminiAPIClient()
-        
-        # 固定最大并发数为 100
-        max_concurrent = 100
-        
-        total_tasks = len(pairs)
-        all_results = []
-        completed = 0
-        success_count = 0
-        fail_count = 0
-        
-        # 计算分批数量
-        num_batches = math.ceil(total_tasks / max_concurrent)
-        
-        # 进度打印配置：任务数 >= 50 时，额外显示百分比里程碑
-        show_milestone = total_tasks >= 50
-        milestones = [0.2, 0.4, 0.6, 0.8, 1.0]  # 20%, 40%, 60%, 80%, 100%
-        milestone_index = 0
-        
-        if num_batches > 1:
-            print(f"BatchNanoBananaPro: 任务数 {total_tasks} 超过并发上限 {max_concurrent}，将分 {num_batches} 批执行")
-        
-        connector = aiohttp.TCPConnector(limit=0, limit_per_host=0)
-        
-        async with aiohttp.ClientSession(connector=connector) as session:
-            for batch_idx in range(num_batches):
-                start_idx = batch_idx * max_concurrent
-                end_idx = min(start_idx + max_concurrent, total_tasks)
-                batch_pairs = pairs[start_idx:end_idx]
-                
-                if num_batches > 1:
-                    print(f"BatchNanoBananaPro: 执行第 {batch_idx + 1}/{num_batches} 批 ({start_idx + 1}-{end_idx})...")
-                
-                # 创建当前批次的任务
-                tasks = []
-                for i, pair in enumerate(batch_pairs):
-                    task = asyncio.create_task(
-                        self._generate_single_task(
-                            client=self.client,
-                            session=session,
-                            prompt=prompt,
-                            model=model,
-                            resolution=resolution,
-                            aspect_ratio=aspect_ratio,
-                            images=list(pair),
-                            output_folder=output_folder,
-                            task_index=start_idx + i
-                        )
-                    )
-                    tasks.append(task)
-                
-                # 使用 as_completed 实时获取完成的任务
-                for coro in asyncio.as_completed(tasks):
-                    result_data = None
-                    try:
-                        result = await coro
-                        if isinstance(result, Exception):
-                            result_data = {
-                                "success": False,
-                                "error": str(result),
-                                "generated_count": 0,
-                                "saved_files": []
-                            }
-                            all_results.append(result_data)
-                        else:
-                            result_data = result
-                            all_results.append(result)
-                    except Exception as e:
-                        result_data = {
-                            "success": False,
-                            "error": str(e),
-                            "generated_count": 0,
-                            "saved_files": []
-                        }
-                        all_results.append(result_data)
-                    
-                    completed += 1
-                    
-                    # 根据成功/失败状态打印不同信息
-                    if result_data and result_data.get("success", False):
-                        success_count += 1
-                        print(f"BatchNanoBananaPro: 任务 {completed}/{total_tasks} 成功 ✓")
-                    else:
-                        fail_count += 1
-                        # 提取错误信息的第一行
-                        error_msg = result_data.get("error", "未知错误") if result_data else "未知错误"
-                        # 截取第一行或前50个字符
-                        if '\n' in error_msg:
-                            error_msg = error_msg.split('\n')[0]
-                        if len(error_msg) > 50:
-                            error_msg = error_msg[:50] + "..."
-                        print(f"BatchNanoBananaPro: 任务 {completed}/{total_tasks} 失败 ✗ - {error_msg}")
-                    
-                    # 更新 ComfyUI 原生进度条
-                    if pbar is not None:
-                        pbar.update(1)
-                    
-                    # 大任务额外显示百分比里程碑
-                    if show_milestone and milestone_index < len(milestones):
-                        progress = completed / total_tasks
-                        if progress >= milestones[milestone_index]:
-                            percentage = int(milestones[milestone_index] * 100)
-                            print(f"BatchNanoBananaPro: >>> 进度 {percentage}% <<<")
-                            milestone_index += 1
-        
-        return all_results
-    
-    def process_batch(
-        self,
-        prompt: str,
-        文件夹1: str,
-        文件夹2: str,
-        文件夹3: str,
-        文件夹4: str,
-        像素缩放: bool,
-        分辨率像素: float,
-        seed: int,
-        保存路径: str,
-        图片配对模式: str,
-        模型: str,
-        宽高比: str,
-        分辨率: str,
-        **kwargs
-    ) -> Tuple[torch.Tensor]:
-        """
-        批量处理图像生成任务
-        
-        Args:
-            prompt: 提示词
-            文件夹1-4: 图片文件夹路径
-            像素缩放: 是否启用像素缩放
-            分辨率像素: 目标像素数（百万像素）
-            seed: 随机种子
-            保存路径: 输出保存路径
-            图片配对模式: 1:1 或 1*N
-            模型: 模型名称
-            宽高比: 输出宽高比
-            分辨率: 输出分辨率
-            **kwargs: 动态参考图输入 (参考图1-9)
-        
-        Returns:
-            输出图像张量
-        """
-        start_time = time.time()
-        
-        try:
-            # 设置随机种子（用于本地随机操作）
-            random.seed(seed)
-            np.random.seed(seed % (2**32))
-            # 验证保存路径
-            if not 保存路径 or not 保存路径.strip():
-                raise ValueError("请提供保存路径")
-            
-            # 加载文件夹图片
-            print("BatchNanoBananaPro: 开始加载图片...")
-            image_lists = self._load_folders(
-                文件夹1, 文件夹2, 文件夹3, 文件夹4,
-                像素缩放, 分辨率像素
-            )
-            
-            # 处理独立的参考图输入
-            manual_images = []
-            for i in range(1, 10):  # 1-9
-                key = f"参考图{i}"
-                if key in kwargs and kwargs[key] is not None:
-                    pil_images = tensor_to_pil(kwargs[key])
-                    for j, img in enumerate(pil_images):
-                        # 如果启用像素缩放，也对参考图进行缩放
-                        if 像素缩放:
-                            img = self.resize_to_megapixels(img, 分辨率像素)
-                        
-                        manual_images.append(
-                            ImageInfo(
-                                image=img,
-                                filename=f"manual_{i}_{j}",
-                                extension=".png",
-                                source_path=""
-                            )
-                        )
-            
-            if manual_images:
-                print(f"BatchNanoBananaPro: 加载了 {len(manual_images)} 张参考图")
-            
-            # 验证是否有图片
-            total_folder_images = sum(len(lst) for lst in image_lists)
-            total_manual_images = len(manual_images)
-            
-            if total_folder_images == 0 and total_manual_images == 0:
-                raise ValueError("未找到任何图片，请检查文件夹路径或提供参考图")
-            
-            # 创建配对
-            print(f"BatchNanoBananaPro: 使用 {图片配对模式} 模式创建配对...")
-            pairs = self._create_pairs(image_lists, 图片配对模式, manual_images if manual_images else None)
-            
-            if not pairs:
-                raise ValueError("配对结果为空，请检查输入")
-            
-            total_tasks = len(pairs)
-            print(f"BatchNanoBananaPro: 共 {total_tasks} 组配对")
-            
-            # 创建 ComfyUI 原生进度条
-            pbar = None
-            if PROGRESS_BAR_AVAILABLE:
-                pbar = ProgressBar(total_tasks)
-            
-            # 初始化 API 客户端
-            if self.client is None:
-                try:
-                    self.client = GeminiAPIClient()
-                except ValueError as e:
-                    raise ValueError(f"初始化 API 客户端失败: {str(e)}")
-            
-            # 执行批量生成
-            print("BatchNanoBananaPro: 开始批量生成...")
-            
-            # 在新线程中运行异步代码，避免事件循环冲突
-            def run_async_in_thread():
-                loop = asyncio.new_event_loop()
-                asyncio.set_event_loop(loop)
-                try:
-                    return loop.run_until_complete(
-                        self._process_batch_async(
-                            pairs=pairs,
-                            prompt=prompt,
-                            model=模型,
-                            resolution=分辨率,
-                            aspect_ratio=宽高比,
-                            output_folder=保存路径,
-                            pbar=pbar
-                        )
-                    )
-                finally:
-                    loop.close()
-            
-            # 使用线程池在新线程中运行事件循环
-            with ThreadPoolExecutor(max_workers=1) as executor:
-                future = executor.submit(run_async_in_thread)
-                results = future.result()
-            
-            # 统计结果
-            success_count = sum(1 for r in results if r.get("success", False))
-            fail_count = len(results) - success_count
-            total_generated = sum(r.get("generated_count", 0) for r in results)
-            all_saved_files = []
-            for r in results:
-                all_saved_files.extend(r.get("saved_files", []))
-            
-            elapsed = time.time() - start_time
-            
-            # 精简统计信息
-            print("=" * 50)
-            print(f"BatchNanoBananaPro 处理完成 | 总耗时: {elapsed:.2f}s | 成功: {success_count}/{total_tasks} | 生成: {total_generated}张")
-            print(f"保存路径: {保存路径}")
-            
-            # 失败详情（如果有）
-            failed_results = [r for r in results if not r.get("success", False)]
-            if failed_results:
-                # 收集失败任务的索引
-                failed_indices = [str(r.get('task_index', '?') + 1) for r in failed_results[:5]]
-                failed_str = ",".join(failed_indices)
-                if len(failed_results) > 5:
-                    failed_str += f"... (共{len(failed_results)}个)"
-                # 显示第一个失败原因作为示例
-                first_error = failed_results[0].get('error', '未知错误')
-                print(f"失败 {len(failed_results)}个: 任务{failed_str} - {first_error}")
-            
-            # 收集所有生成的图片
-            output_images = []
-            for file_path in all_saved_files:
-                try:
-                    img = Image.open(file_path)
-                    output_images.append(img)
-                except Exception as e:
-                    print(f"BatchNanoBananaPro: 无法加载图片 {file_path} - {e}")
-            
-            # 如果没有生成成功的图片，创建一个占位图
-            if not output_images:
-                placeholder = Image.new('RGB', (512, 512), color=(128, 128, 128))
-                output_images = [placeholder]
-            
-            # 转换为张量
-            output_tensor = pil_to_tensor(output_images)
-            
-            return (output_tensor,)
-        
-        except ValueError as e:
-            # 检测是否为授权错误
-            if str(e) == "未授权！":
-                print("请联系作者授权后方可使用！")
-            else:
-                print(f"BatchNanoBananaPro: 输入错误 - {str(e)}")
-            raise
-        
-        except RuntimeError as e:
-            print(f"BatchNanoBananaPro: 运行时错误 - {str(e)}")
-            raise
-        
-        except Exception as e:
-            print(f"BatchNanoBananaPro: 未知错误 - {str(e)}")
-            raise
-        
-        finally:
-            # 无论成功或失败，都尝试查询余额
-            if self.client is not None:
-                try:
-                    balance_data = self.client.query_balance_sync()
-                    balance_info = self.client.format_balance_info(balance_data)
-                    print(f"{balance_info}")
-                    print("=" * 50)
-                except Exception as e:
-                    print(f"⚠️ 余额查询失败 - {str(e)}")
-                    print("=" * 50)
@@ -0,0 +1,93 @@
+"""
+o1key 颜色去背景节点
+基于颜色距离计算，精确可控，不依赖 AI 模型
+"""
+
+import numpy as np
+import torch
+from PIL import Image
+
+
+class O1keyColorRemoveBG:
+    """
+    颜色去背景 - 精确移除纯色背景
+
+    模式说明：
+      - 白色(white): 移除白色背景，适合大多数场景
+      - 白色保护(white-preserve): 移除白底但保护浅色前景物体
+      - 自动检测(corner): 自动采样四角颜色作为背景色
+      - 指定颜色(color): 手动指定要移除的背景颜色
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+                "模式": (["白色", "白色保护", "自动检测", "指定颜色"], {
+                    "default": "白色",
+                }),
+                "容差": ("FLOAT", {
+                    "default": 8.0,
+                    "min": 0.0,
+                    "max": 100.0,
+                    "step": 1.0,
+                    "tooltip": "颜色距离阈值，越大去除范围越广",
+                }),
+                "羽化": ("FLOAT", {
+                    "default": 45.0,
+                    "min": 0.0,
+                    "max": 200.0,
+                    "step": 1.0,
+                    "tooltip": "边缘过渡范围，越大边缘越柔和",
+                }),
+            },
+            "optional": {
+                "背景色R": ("INT", {"default": 255, "min": 0, "max": 255}),
+                "背景色G": ("INT", {"default": 255, "min": 0, "max": 255}),
+                "背景色B": ("INT", {"default": 255, "min": 0, "max": 255}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("RGBA图像",)
+    FUNCTION = "remove_bg"
+    CATEGORY = "o1key/image"
+
+    _MODE_MAP = {
+        "白色": "white",
+        "白色保护": "white-preserve",
+        "自动检测": "corner",
+        "指定颜色": "color",
+    }
+
+    def remove_bg(self, image, 模式, 容差, 羽化, 背景色R=255, 背景色G=255, 背景色B=255):
+        from ..utils.color_key import remove_background
+
+        mode = self._MODE_MAP.get(模式, "white")
+        bg_color = (背景色R, 背景色G, 背景色B)
+
+        batch_size = image.shape[0]
+        results = []
+
+        for i in range(batch_size):
+            frame = image[i]  # [H, W, C]
+            arr = (frame.cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
+
+            if arr.shape[2] == 4:
+                pil_img = Image.fromarray(arr, mode="RGBA")
+            else:
+                pil_img = Image.fromarray(arr, mode="RGB")
+
+            result = remove_background(
+                pil_img, mode=mode, bg_color=bg_color,
+                tolerance=容差, feather=羽化,
+            )
+
+            result_arr = np.array(result.convert("RGBA")).astype(np.float32) / 255.0
+            results.append(torch.from_numpy(result_arr))
+
+        output = torch.stack(results, dim=0)
+        print(f"[o1key 颜色去背景] 模式={模式}, 容差={容差}, 羽化={羽化}, "
+              f"处理 {batch_size} 张")
+        return (output,)
@@ -0,0 +1,420 @@
+"""
+豆包生图节点
+后端通过 new-api 兼容层调用豆包官方 API
+"""
+
+import asyncio
+import time
+import numpy as np
+import torch
+from concurrent.futures import ThreadPoolExecutor
+from PIL import Image
+from typing import List, Optional
+
+from ..clients.doubao_image_client import DoubaoImageClient
+from ..utils.image_utils import tensor_to_pil
+
+
+# ── 模型列表 ──────────────────────────────────────────────────────────────────
+_MODELS = [
+    "doubao-seedream-5-0-260128",
+    "doubao-seedream-4-5-251128",
+]
+
+# ── 宽高比列表 ─────────────────────────────────────────────────────────────────
+_ASPECT_RATIOS = ["1:1", "4:3", "3:4", "16:9", "9:16", "3:2", "2:3", "21:9"]
+
+# ── 分辨率档位（每个模型支持的档位不同）──────────────────────────────────────
+# 5.0：2K / 3K
+# 4.5：2K / 4K
+_RESOLUTIONS = ["2K", "3K", "4K"]
+
+# ── 像素对照表 ─────────────────────────────────────────────────────────────────
+# 结构：{ 模型版本key: { 分辨率: { 宽高比: (宽, 高) } } }
+_SIZE_TABLE = {
+    "5-0": {
+        "2K": {
+            "1:1":  (2048, 2048),
+            "4:3":  (2304, 1728),
+            "3:4":  (1728, 2304),
+            "16:9": (2848, 1600),
+            "9:16": (1600, 2848),
+            "3:2":  (2496, 1664),
+            "2:3":  (1664, 2496),
+            "21:9": (3136, 1344),
+        },
+        "3K": {
+            "1:1":  (3072, 3072),
+            "4:3":  (3456, 2592),
+            "3:4":  (2592, 3456),
+            "16:9": (4096, 2304),
+            "9:16": (2304, 4096),
+            "3:2":  (3744, 2496),
+            "2:3":  (2496, 3744),
+            "21:9": (4704, 2016),
+        },
+    },
+    "4-5": {
+        "2K": {
+            "1:1":  (2048, 2048),
+            "4:3":  (2304, 1728),
+            "3:4":  (1728, 2304),
+            "16:9": (2848, 1600),
+            "9:16": (1600, 2848),
+            "3:2":  (2496, 1664),
+            "2:3":  (1664, 2496),
+            "21:9": (3136, 1344),
+        },
+        "4K": {
+            "1:1":  (4096, 4096),
+            "4:3":  (4704, 3520),
+            "3:4":  (3520, 4704),
+            "16:9": (5504, 3040),
+            "9:16": (3040, 5504),
+            "3:2":  (4992, 3328),
+            "2:3":  (3328, 4992),
+            "21:9": (6240, 2656),
+        },
+    },
+}
+
+# 每个模型版本支持的分辨率档位
+_MODEL_RESOLUTIONS = {
+    "5-0": ["2K", "3K"],
+    "4-5": ["2K", "4K"],
+}
+
+# 并发请求超时（秒）
+_CONCURRENT_TIMEOUT = 330
+
+
+def _model_key(model: str) -> str:
+    """从模型 ID 中提取版本 key（'5-0' 或 '4-5'）。"""
+    for key in _SIZE_TABLE:
+        if key in model:
+            return key
+    raise ValueError(f"无法识别模型版本：{model}，支持的模型：{_MODELS}")
+
+
+def _pil_list_to_tensor(images: List[Image.Image]) -> torch.Tensor:
+    """
+    PIL Image 列表 → ComfyUI IMAGE tensor [B, H, W, C]，值域 [0, 1]。
+    多张尺寸不同时，以最大尺寸为准，较小图像丢弃。
+    """
+    if not images:
+        placeholder = Image.new("RGB", (512, 512), color=(128, 128, 128))
+        images = [placeholder]
+
+    base_size = max(images, key=lambda img: img.size[0] * img.size[1]).size
+    matched = [img for img in images if img.size == base_size]
+    skipped = len(images) - len(matched)
+    if skipped:
+        print(f"[豆包生图] 丢弃 {skipped} 张非最大尺寸图像，仅输出 {base_size[0]}×{base_size[1]} 的 {len(matched)} 张")
+
+    tensors = []
+    for img in matched:
+        arr = np.array(img.convert("RGB")).astype(np.float32) / 255.0
+        tensors.append(torch.from_numpy(arr))
+
+    return torch.stack(tensors, dim=0)  # [B, H, W, C]
+
+
+class DoubaoImage:
+    """豆包生图 —— 通过宽高比 + 分辨率档位选择尺寸，后端自动换算真实像素"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "模型": (
+                    _MODELS,
+                    {"default": _MODELS[0]},
+                ),
+                "提示词": (
+                    "STRING",
+                    {
+                        "multiline": True,
+                        "default": "",
+                        "tooltip": "用于创建或编辑图像的文本提示",
+                    },
+                ),
+                "宽高比": (
+                    _ASPECT_RATIOS,
+                    {
+                        "default": "1:1",
+                        "tooltip": "图像宽高比。所有分辨率档位均支持这些比例",
+                    },
+                ),
+                "分辨率": (
+                    _RESOLUTIONS,
+                    {
+                        "default": "2K",
+                        "tooltip": (
+                            "图像分辨率档位。\n"
+                            "• Seedream 5.0：支持 2K / 3K\n"
+                            "• Seedream 4.5：支持 2K / 4K\n"
+                            "（3K 与 4.5 或 4K 与 5.0 搭配时将报错）"
+                        ),
+                    },
+                ),
+                "生图数量": (
+                    "INT",
+                    {
+                        "default": 1,
+                        "min": 1,
+                        "max": 10,
+                        "step": 1,
+                        "tooltip": "生成图像的数量。2-10 张时自动并发请求，加快出图速度",
+                    },
+                ),
+                "种子": (
+                    "INT",
+                    {
+                        "default": 0,
+                        "min": 0,
+                        "max": 2147483647,
+                        "step": 1,
+                        "control_after_generate": True,
+                        "tooltip": "用于生成的随机种子",
+                    },
+                ),
+                "部分失败时停止": (
+                    "BOOLEAN",
+                    {
+                        "default": True,
+                        "tooltip": (
+                            "启用时：任意一张失败即抛出错误并中止。\n"
+                            "禁用时：返回已成功生成的图像，忽略失败项"
+                        ),
+                    },
+                ),
+            },
+            "optional": {
+                "图像": (
+                    "IMAGE",
+                    {
+                        "tooltip": (
+                            "用于图生图的输入图像。"
+                            "单参考或多参考生成时，可输入1-10张图像列表"
+                        ),
+                    },
+                ),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("图像",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/豆包"
+
+    # ── 并发核心：在新 event loop 里 gather N 个 _generate_async ─────────────
+
+    async def _run_concurrent(
+        self,
+        client: DoubaoImageClient,
+        生图数量: int,
+        model: str,
+        prompt: str,
+        size: str,
+        seed: int,
+        image_tensor,
+        pbar,
+    ) -> List[dict]:
+        """
+        并发发起 生图数量 个独立请求，每完成一个推进一格进度条。
+        返回结果列表：[{"index": int, "images": [...], "error": str|None}]
+        """
+        # 固定参数（顺序生成功能暂时隐藏）
+        seq = "disabled"
+        max_img = 1
+
+        async def _one(idx: int) -> dict:
+            try:
+                imgs = await client._generate_async(
+                    model=model,
+                    prompt=prompt,
+                    size=size,
+                    seed=seed,
+                    sequential_image_generation=seq,
+                    max_images=max_img,
+                    image_tensor=image_tensor,
+                )
+                return {"index": idx, "images": imgs, "error": None}
+            except Exception as e:
+                return {"index": idx, "images": [], "error": str(e)}
+
+        # 用 as_completed 方式逐个推进进度条
+        tasks = [asyncio.create_task(_one(i)) for i in range(生图数量)]
+        results = [None] * 生图数量
+        completed = 0
+
+        for coro in asyncio.as_completed(tasks):
+            res = await coro
+            results[res["index"]] = res
+            completed += 1
+            status = "✓" if res["error"] is None else f"✗ {res['error']}"
+            print(f"[豆包生图] [{completed}/{生图数量}] 第 {res['index'] + 1} 张 → {status}")
+            if pbar is not None:
+                pbar.update(1)
+
+        return results
+
+    # ── 节点主入口 ────────────────────────────────────────────────────────────
+
+    def generate(
+        self,
+        模型: str,
+        提示词: str,
+        宽高比: str,
+        分辨率: str,
+        生图数量: int,
+        种子: int,
+        部分失败时停止: bool,
+        图像=None,
+    ):
+        start_time = time.time()
+
+        # 顺序图像生成功能暂时隐藏，固定使用默认值
+        顺序图像生成 = "disabled"
+        最大图片数 = 1
+
+        # ── 1. 校验提示词 ─────────────────────────────────────────────────────
+        if not 提示词.strip():
+            raise ValueError("提示词不能为空，请输入图像描述后重试。")
+
+        # ── 2. 解析模型版本并校验分辨率兼容性 ────────────────────────────────
+        try:
+            mkey = _model_key(模型)
+        except ValueError as e:
+            raise ValueError(str(e)) from None
+
+        supported = _MODEL_RESOLUTIONS[mkey]
+        if 分辨率 not in supported:
+            raise ValueError(
+                f"模型 {模型} 不支持 {分辨率} 分辨率。\n"
+                f"该模型支持：{' / '.join(supported)}"
+            )
+
+        # ── 3. 查表换算真实像素 ───────────────────────────────────────────────
+        w, h = _SIZE_TABLE[mkey][分辨率][宽高比]
+        size_str = f"{w}x{h}"
+
+        # ── 4. 打印概要 ───────────────────────────────────────────────────────
+        mode_str = "图生图" if 图像 is not None else "文生图"
+        print(
+            f"[豆包生图] {mode_str} | 模型={模型} | {分辨率} {宽高比} → {size_str}"
+            f" | 数量={生图数量} | 种子={种子}"
+        )
+
+        # ── 5. 初始化客户端 ───────────────────────────────────────────────────
+        try:
+            client = DoubaoImageClient()
+        except ValueError as e:
+            raise ValueError(str(e)) from None
+
+        # ── 6. 进度条（按张数计）──────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(生图数量)
+        except Exception:
+            pbar = None
+
+        # ── 7. 单张 / 多张分支 ────────────────────────────────────────────────
+        if 生图数量 == 1:
+            # 单张：走原有同步路径
+            try:
+                pil_images: List[Image.Image] = client.generate_sync(
+                    model=模型,
+                    prompt=提示词,
+                    size=size_str,
+                    seed=种子,
+                    sequential_image_generation=顺序图像生成,
+                    max_images=最大图片数,
+                    image_tensor=图像,
+                )
+            except RuntimeError as e:
+                raise RuntimeError(str(e)) from None
+            except Exception as e:
+                raise RuntimeError(f"豆包生图请求失败: {e}") from None
+
+            if pbar is not None:
+                pbar.update(1)
+
+        else:
+            # 多张：并发请求
+            def _run_in_thread():
+                loop = asyncio.new_event_loop()
+                asyncio.set_event_loop(loop)
+                try:
+                    return loop.run_until_complete(
+                        self._run_concurrent(
+                            client=client,
+                            生图数量=生图数量,
+                            model=模型,
+                            prompt=提示词,
+                            size=size_str,
+                            seed=种子,
+                            image_tensor=图像,
+                            pbar=pbar,
+                        )
+                    )
+                finally:
+                    loop.close()
+
+            with ThreadPoolExecutor(max_workers=1) as executor:
+                future = executor.submit(_run_in_thread)
+                try:
+                    results = future.result(timeout=_CONCURRENT_TIMEOUT)
+                except TimeoutError:
+                    raise RuntimeError(
+                        f"并发生图超时（>{_CONCURRENT_TIMEOUT}s），请检查网络或减少生图数量"
+                    )
+
+            # 统计成功 / 失败
+            success_results = [r for r in results if r and r["error"] is None]
+            failed_results  = [r for r in results if r and r["error"] is not None]
+
+            if failed_results:
+                fail_info = "；".join(
+                    f"第{r['index']+1}张: {r['error']}" for r in failed_results
+                )
+                if 部分失败时停止:
+                    raise RuntimeError(
+                        f"{len(failed_results)}/{生图数量} 张生成失败：{fail_info}\n"
+                        "（可将【部分失败时停止】设为 False 以返回已成功的图像）"
+                    )
+                else:
+                    print(f"[豆包生图] 警告：{len(failed_results)}/{生图数量} 张失败，已忽略：{fail_info}")
+
+            if not success_results:
+                raise RuntimeError("所有图像均生成失败，请检查网络或 API 配置。")
+
+            # 按原始 index 排序，展平为 PIL 列表
+            success_results.sort(key=lambda r: r["index"])
+            pil_images = []
+            for r in success_results:
+                pil_images.extend(r["images"])
+
+        # ── 8. PIL → tensor ───────────────────────────────────────────────────
+        output_tensor = _pil_list_to_tensor(pil_images)
+
+        # ── 9. 完成日志 ───────────────────────────────────────────────────────
+        elapsed = time.time() - start_time
+        print(
+            f"[豆包生图] 完成！耗时 {elapsed:.1f}s，"
+            f"输出 {output_tensor.shape[0]} 张 "
+            f"{output_tensor.shape[2]}×{output_tensor.shape[1]}"
+        )
+
+        return (output_tensor,)
+
+
+# ── 节点注册 ──────────────────────────────────────────────────────────────────
+
+NODE_CLASS_MAPPINGS = {
+    "DoubaoImage": DoubaoImage,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "DoubaoImage": "豆包生图",
+}
@@ -0,0 +1,170 @@
+"""
+Flux2 图像编辑节点
+通过 vip.o1key.com 调用 Flux2 + SeedVR2 远程服务进行图像编辑和超分辨率
+
+功能：
+- 接收主图和参考图
+- 上传到远程服务器执行图像编辑
+- 轮询等待 SeedVR2 超分辨率结果
+- 返回最终放大后的图像
+"""
+
+import time
+from io import BytesIO
+from typing import Tuple
+
+import torch
+from PIL import Image
+
+from ..utils.image_utils import tensor_to_pil, pil_to_tensor
+from ..clients.flux_edit_client import FluxEditClient
+
+
+class FluxImageEdit:
+    """
+    Flux2 图像编辑节点
+
+    通过远程 API 将主图与参考图结合，按照提示词进行图像编辑，
+    并经 SeedVR2 超分辨率放大后返回最终结果。
+    """
+
+    SIZES = ["2K", "4K"]
+
+    def __init__(self):
+        self.client = None
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "主图": ("IMAGE",),
+                "参考图": ("IMAGE",),
+                "提示词": ("STRING", {
+                    "default": "Replace the woman's underwear in Figure 1 with the strapless bra in Figure 2",
+                    "multiline": True,
+                }),
+                "分辨率": (cls.SIZES, {
+                    "default": "4K",
+                }),
+                "轮询间隔": ("INT", {
+                    "default": 15,
+                    "min": 5,
+                    "max": 60,
+                    "step": 5,
+                }),
+                "seed": ("INT", {
+                    "default": 0,
+                    "min": 0,
+                    "max": 0xffffffffffffffff,
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("输出图像",)
+    FUNCTION = "generate"
+    CATEGORY = "image/edit"
+
+    def _image_to_jpeg_bytes(self, image: Image.Image, quality: int = 92) -> bytes:
+        """将 PIL Image 转为 JPEG 二进制"""
+        if image.mode in ("RGBA", "P", "LA"):
+            image = image.convert("RGB")
+        buf = BytesIO()
+        image.save(buf, format="JPEG", quality=quality)
+        return buf.getvalue()
+
+    def generate(
+        self,
+        主图: torch.Tensor,
+        参考图: torch.Tensor,
+        提示词: str,
+        分辨率: str,
+        轮询间隔: int,
+        seed: int,
+    ) -> Tuple[torch.Tensor]:
+        """
+        执行图像编辑
+
+        Args:
+            主图: 要编辑的原始图像 (ComfyUI tensor, [B, H, W, C])
+            参考图: 参考/风格图像 (ComfyUI tensor, [B, H, W, C])
+            提示词: 编辑指令
+            分辨率: 超分辨率目标 ("2K" 或 "4K"，会自动映射为 2048/4096)
+            轮询间隔: 轮询秒数
+            seed: 随机种子
+
+        Returns:
+            输出图像 tensor (IMAGE,)
+        """
+        start_time = time.time()
+
+        try:
+            # 初始化客户端
+            if self.client is None:
+                self.client = FluxEditClient()
+
+            # Tensor → PIL（取第一张）
+            main_pils = tensor_to_pil(主图)
+            ref_pils = tensor_to_pil(参考图)
+
+            if not main_pils:
+                raise ValueError("主图不能为空")
+            if not ref_pils:
+                raise ValueError("参考图不能为空")
+
+            main_img = main_pils[0]
+            ref_img = ref_pils[0]
+
+            # PIL → JPEG bytes
+            main_bytes = self._image_to_jpeg_bytes(main_img)
+            ref_bytes = self._image_to_jpeg_bytes(ref_img)
+
+            print(f"Flux Edit: 开始处理 | 主图 {main_img.size} | 参考图 {ref_img.size} | 分辨率 {分辨率} | seed {seed}")
+
+            # 进度回调
+            def progress_callback(status_str: str):
+                print(f"Flux Edit: {status_str}")
+
+            # 提交任务并等待结果
+            result_bytes = self.client.submit_and_wait(
+                image_bytes=main_bytes,
+                mask_bytes=ref_bytes,
+                prompt=提示词,
+                size=分辨率,
+                poll_interval=轮询间隔,
+                progress_callback=progress_callback,
+            )
+
+            # 解码结果
+            result_img = Image.open(BytesIO(result_bytes))
+            if result_img.mode != "RGB":
+                result_img = result_img.convert("RGB")
+
+            print(f"Flux Edit: 结果图像尺寸 {result_img.size}")
+
+            # 转为 tensor
+            output_tensor = pil_to_tensor([result_img])
+
+            # 打印耗时
+            elapsed = time.time() - start_time
+            if elapsed < 60:
+                time_str = f"{elapsed:.1f}s"
+            else:
+                minutes = int(elapsed // 60)
+                seconds = elapsed % 60
+                time_str = f"{minutes}m {seconds:.0f}s"
+            print(f"Flux Edit: 完成！总耗时 {time_str}")
+
+            return (output_tensor,)
+
+        except ValueError as e:
+            if str(e) == "未授权！":
+                print("请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            print(f"Flux Edit: ❌ {e}")
+            raise
+
+        except Exception as e:
+            error_msg = str(e)
+            print(f"Flux Edit: ❌ {error_msg}")
+            raise RuntimeError(error_msg) from None
@@ -6,29 +6,49 @@ ComfyUI 自定义节点,用于调用 Gemini Flash 模型进行多模态文本生
 import base64
 import os
 import time
+import tempfile
 from typing import Dict, List, Optional, Tuple
+from io import BytesIO

 import torch
+from PIL import Image

 from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.file_types import FileData
 from ..clients.gemini_flash_client import GeminiFlashClient
 from ..models_config import get_enabled_flash_models

+# 文件大小限制（20MB）
+MAX_FILE_SIZE = 20 * 1024 * 1024
+
+# 图片缩放后最大尺寸（1K分辨率 = 1024像素）
+MAX_IMAGE_DIMENSION = 1024
+
+# 视频压缩目标大小（1-10MB）
+TARGET_VIDEO_SIZE_MIN = 1 * 1024 * 1024
+TARGET_VIDEO_SIZE_MAX = 10 * 1024 * 1024
+

 # 支持的视频 MIME 类型映射
 VIDEO_MIME_TYPES = {
    ".mp4": "video/mp4",
    ".mpeg": "video/mpeg",
    ".mpg": "video/mpg",
-    ".mov": "video/mov",
-    ".avi": "video/avi",
+    ".mov": "video/quicktime",
+    ".avi": "video/x-msvideo",
    ".flv": "video/x-flv",
    ".webm": "video/webm",
-    ".wmv": "video/wmv",
+    ".wmv": "video/x-ms-wmv",
    ".3gp": "video/3gpp",
    ".3gpp": "video/3gpp"
 }

+try:
+    import subprocess
+    FFMPEG_AVAILABLE = True
+except ImportError:
+    FFMPEG_AVAILABLE = False
+

 class GoogleGemini:
    """
@@ -36,14 +56,13 @@ class GoogleGemini:
    
    功能：
    - 支持多个 Gemini Flash 模型
-    - 支持图片和视频输入
-    - 支持系统指令
-    - 支持不同思考深度（不思考/高）
-    - 输出生成的文本内容
+    - 支持图片、视频和文件输入
+    - 支持不同思考等级（不思考/低/中/高）- 通过 thinkingConfig.thinkingLevel 控制
+    - 输出生成的文本内容（主要内容 + 思考内容）
    """
    
-    # 支持的思考深度选项
-    THINKING_DEPTHS = ["不思考", "高"]
+    # 支持的思考等级选项
+    THINKING_LEVELS = ["不思考", "低", "中", "高"]
    
    def __init__(self):
        """初始化节点"""
@@ -67,23 +86,20 @@ class GoogleGemini:
                    "default": "",
                    "multiline": True
                }),
-                "思考深度": (cls.THINKING_DEPTHS, {
+                "思考等级": (cls.THINKING_LEVELS, {
                    "default": "不思考"
                })
            },
            "optional": {
-                "系统指令": ("STRING", {
-                    "default": "",
-                    "multiline": True
-                }),
                "图片": ("IMAGE",),
-                "视频": ("VIDEO",)
+                "视频": ("VIDEO",),
+                "文件": ("FILE",)
            }
        }
    
    # 返回值类型
-    RETURN_TYPES = ("STRING", "STRING")
-    RETURN_NAMES = ("主要内容", "思考内容")
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("主要内容",)
    
    # 执行函数名
    FUNCTION = "generate"
@@ -94,6 +110,66 @@ class GoogleGemini:
    # 允许输出到 UI
    OUTPUT_NODE = True
    
+    def _resize_image_if_needed(self, img: Image.Image) -> Image.Image:
+        """
+        如果图片过大，缩放到1K分辨率
+        
+        Args:
+            img: PIL Image 对象
+            
+        Returns:
+            缩放后的 PIL Image
+        """
+        width, height = img.size
+        max_dim = max(width, height)
+        
+        if max_dim > MAX_IMAGE_DIMENSION:
+            # 计算缩放比例
+            scale = MAX_IMAGE_DIMENSION / max_dim
+            new_width = int(width * scale)
+            new_height = int(height * scale)
+            
+            print(f"Google Gemini: 图片尺寸 {width}x{height} 超过限制，缩放至 {new_width}x{new_height}")
+            img = img.resize((new_width, new_height), Image.Resampling.LANCZOS)
+        
+        return img
+    
+    def _check_and_compress_image(self, img: Image.Image) -> str:
+        """
+        检查图片大小，如果超过20MB则进行压缩
+        
+        Args:
+            img: PIL Image 对象
+            
+        Returns:
+            base64 编码的字符串
+        """
+        # 先进行尺寸缩放（如果需要）
+        img = self._resize_image_if_needed(img)
+        
+        # 尝试不同的压缩质量
+        qualities = [95, 85, 75, 65, 55, 45]
+        
+        for quality in qualities:
+            buffer = BytesIO()
+            # 转换为RGB模式（去除alpha通道）以减小体积
+            if img.mode in ('RGBA', 'P'):
+                img_rgb = img.convert('RGB')
+            else:
+                img_rgb = img
+            
+            img_rgb.save(buffer, format='JPEG', quality=quality, optimize=True)
+            buffer.seek(0)
+            data = buffer.getvalue()
+            
+            if len(data) <= MAX_FILE_SIZE:
+                print(f"Google Gemini: 图片压缩后大小 {len(data) / 1024 / 1024:.2f}MB (质量{quality})")
+                return base64.b64encode(data).decode('utf-8')
+        
+        # 如果所有质量都无法满足，使用最低质量
+        print(f"Google Gemini: 警告 - 即使最低质量仍超过20MB，将使用最低质量发送")
+        return base64.b64encode(data).decode('utf-8')
+    
    def _prepare_image_data(
        self, 
        images: Optional[torch.Tensor]
@@ -101,6 +177,8 @@ class GoogleGemini:
        """
        准备图片数据
        
+        如果图片超过20MB，会自动进行缩放和压缩
+        
        Args:
            images: ComfyUI 图片张量 [B, H, W, C]
        
@@ -110,17 +188,183 @@ class GoogleGemini:
        if images is None:
            return None
        
-        image_data = []
        pil_images = tensor_to_pil(images)
+        if not pil_images:
+            return None
        
+        # 将所有图片转为 RGB PIL Image 并首次编码
+        processed = []  # [(pil_img_rgb, b64_data, mime_type)]
        for img in pil_images:
-            b64_str = encode_image_to_base64(img)
-            image_data.append({
-                "mime_type": "image/png",
-                "data": b64_str
-            })
+            buffer = BytesIO()
+            img.save(buffer, format='PNG')
+            original_size = buffer.tell()
+            buffer.close()
+            
+            if original_size > MAX_FILE_SIZE:
+                print(f"Google Gemini: 检测到图片过大 ({original_size / 1024 / 1024:.2f}MB)，正在进行压缩...")
+                img_rgb = img.convert('RGB') if img.mode != 'RGB' else img.copy()
+                b64_str = self._check_and_compress_image(img_rgb)
+                processed.append((img_rgb, b64_str, "image/jpeg"))
+            else:
+                b64_str = encode_image_to_base64(img)
+                processed.append((None, b64_str, "image/png"))
        
-        return image_data if image_data else None
+        # 多图总体积控制
+        def calc_total_bytes():
+            return sum(len(base64.b64decode(item[1])) for item in processed)
+        
+        total = calc_total_bytes()
+        if total > MAX_FILE_SIZE and len(processed) > 1:
+            print(f"Google Gemini: 图片总体积 {total / 1024 / 1024:.2f}MB 超过 {MAX_FILE_SIZE // 1024 // 1024}MB 限制，正在压缩...")
+            
+            # 降质量
+            for quality in range(70, 19, -10):
+                new_processed = []
+                for pil_img, _, _ in processed:
+                    if pil_img is None:
+                        # PNG 原图需要转 RGB
+                        continue
+                    buf = BytesIO()
+                    pil_img.save(buf, format='JPEG', quality=quality, optimize=True)
+                    data = buf.getvalue()
+                    new_processed.append((pil_img, base64.b64encode(data).decode('utf-8'), "image/jpeg"))
+                if not new_processed:
+                    break
+                processed = new_processed
+                total = calc_total_bytes()
+                if total <= MAX_FILE_SIZE:
+                    print(f"Google Gemini: 图片压缩完成，总体积 {total / 1024 / 1024:.2f}MB ({len(processed)}张图片，质量{quality})")
+                    break
+            
+            # 降分辨率
+            if total > MAX_FILE_SIZE:
+                for scale in [0.75, 0.5, 0.35]:
+                    new_processed = []
+                    for pil_img, _, _ in processed:
+                        if pil_img is None:
+                            continue
+                        w, h = pil_img.size
+                        resized = pil_img.resize((int(w * scale), int(h * scale)), Image.Resampling.LANCZOS)
+                        buf = BytesIO()
+                        resized.save(buf, format='JPEG', quality=20, optimize=True)
+                        data = buf.getvalue()
+                        new_processed.append((resized, base64.b64encode(data).decode('utf-8'), "image/jpeg"))
+                    if not new_processed:
+                        break
+                    processed = new_processed
+                    total = calc_total_bytes()
+                    if total <= MAX_FILE_SIZE:
+                        print(f"Google Gemini: 图片压缩完成，总体积 {total / 1024 / 1024:.2f}MB ({len(processed)}张图片，缩放{int(scale*100)}%)")
+                        break
+            
+            if total > MAX_FILE_SIZE:
+                print(f"Google Gemini: 无法将 {len(processed)} 张图片压缩到 {MAX_FILE_SIZE // 1024 // 1024}MB 以内，请减少图片数量或降低分辨率")
+                raise ValueError(f"图片总体积 {total / 1024 / 1024:.2f}MB 超过限制，无法压缩到 {MAX_FILE_SIZE // 1024 // 1024}MB 以内")
+        
+        image_data = [{"mime_type": mt, "data": b64} for _, b64, mt in processed]
+        return image_data
+    
+    def _compress_video_with_ffmpeg(self, input_path: str, output_path: str, target_size: int) -> bool:
+        """
+        使用 FFmpeg 压缩视频到目标大小
+        
+        Args:
+            input_path: 输入视频路径
+            output_path: 输出视频路径
+            target_size: 目标文件大小（字节）
+            
+        Returns:
+            是否压缩成功
+        """
+        try:
+            # 获取视频时长（秒）
+            probe_cmd = ['ffprobe', '-v', 'error', '-show_entries', 'format=duration', 
+                        '-of', 'default=noprint_wrappers=1:nokey=1', input_path]
+            duration = float(subprocess.check_output(probe_cmd).decode().strip())
+            
+            # 计算目标比特率（bit/s），预留一些余量
+            target_bitrate = int((target_size * 8) / duration * 0.9)
+            
+            # 使用 FFmpeg 压缩视频
+            # -c:v libx264: 使用 H.264 编码器
+            # -b:v: 视频比特率
+            # -maxrate 和 -bufsize: 控制码率波动
+            # -c:a aac: 音频使用 AAC 编码
+            # -b:a 128k: 音频比特率 128k
+            # -movflags +faststart: 优化网络播放
+            cmd = [
+                'ffmpeg', '-y', '-i', input_path,
+                '-c:v', 'libx264',
+                '-b:v', f'{target_bitrate}',
+                '-maxrate', f'{int(target_bitrate * 1.5)}',
+                '-bufsize', f'{target_bitrate * 2}',
+                '-c:a', 'aac',
+                '-b:a', '128k',
+                '-movflags', '+faststart',
+                '-preset', 'fast',
+                output_path
+            ]
+            
+            print(f"Google Gemini: 正在压缩视频到 {target_size / 1024 / 1024:.1f}MB...")
+            result = subprocess.run(cmd, capture_output=True, text=True)
+            
+            if result.returncode == 0 and os.path.exists(output_path):
+                final_size = os.path.getsize(output_path)
+                print(f"Google Gemini: 视频压缩完成，最终大小 {final_size / 1024 / 1024:.2f}MB")
+                return True
+            else:
+                print(f"Google Gemini: FFmpeg 压缩失败: {result.stderr}")
+                return False
+                
+        except Exception as e:
+            print(f"Google Gemini: 视频压缩异常: {str(e)}")
+            return False
+    
+    def _compress_video(self, video_path: str) -> str:
+        """
+        压缩视频到 1-10MB 之间
+        
+        Args:
+            video_path: 原视频路径
+            
+        Returns:
+            压缩后的视频路径（临时文件）
+        """
+        original_size = os.path.getsize(video_path)
+        print(f"Google Gemini: 视频文件过大 ({original_size / 1024 / 1024:.2f}MB)，正在压缩...")
+        
+        # 创建临时文件
+        temp_dir = tempfile.gettempdir()
+        _, ext = os.path.splitext(video_path)
+        output_path = os.path.join(temp_dir, f"compressed_{int(time.time())}{ext}")
+        
+        # 确定目标大小（优先尝试 10MB，如果不行再降低）
+        target_sizes = [
+            TARGET_VIDEO_SIZE_MAX,  # 10MB
+            int(TARGET_VIDEO_SIZE_MAX * 0.8),  # 8MB
+            int(TARGET_VIDEO_SIZE_MAX * 0.6),  # 6MB
+            int(TARGET_VIDEO_SIZE_MAX * 0.5),  # 5MB
+            TARGET_VIDEO_SIZE_MIN * 5,  # 5MB
+            TARGET_VIDEO_SIZE_MIN * 3,  # 3MB
+            TARGET_VIDEO_SIZE_MIN * 2,  # 2MB
+        ]
+        
+        # 优先尝试 FFmpeg
+        if FFMPEG_AVAILABLE:
+            for target_size in target_sizes:
+                if self._compress_video_with_ffmpeg(video_path, output_path, target_size):
+                    # 检查最终大小
+                    final_size = os.path.getsize(output_path)
+                    if TARGET_VIDEO_SIZE_MIN <= final_size <= MAX_FILE_SIZE:
+                        return output_path
+                    # 如果仍然太大，继续降低目标
+                    os.remove(output_path)
+
+        # 所有压缩方法都失败
+        raise ValueError(
+            f"视频文件过大 ({original_size / 1024 / 1024:.2f}MB) 且无法压缩到 20MB 以下。"
+            f"请安装 FFmpeg 以获得更好的压缩效果，或手动压缩视频。"
+        )
    
    def _prepare_video_data(
        self, 
@@ -131,6 +375,7 @@ class GoogleGemini:
        
        ComfyUI VIDEO 类型包含视频文件路径信息。
        读取视频文件并转换为 base64。
+        如果视频超过 20MB，会自动进行压缩。
        
        Args:
            video: ComfyUI VIDEO 类型数据
@@ -141,17 +386,48 @@ class GoogleGemini:
        if video is None:
            return None
        
-        # VIDEO 类型通常是一个字典，包含 'video' 键指向文件路径
-        # 或者直接是文件路径字符串
+        # VIDEO 类型处理：支持多种格式
        video_path = None
+        temp_compressed_path = None
        
        if isinstance(video, dict):
-            # 尝试获取视频路径
-            video_path = video.get("video") or video.get("path") or video.get("file")
+            # 字典格式：尝试常见的键名
+            video_path = video.get("video") or video.get("path") or video.get("file") or video.get("filename")
+            # 如果还是找不到，遍历所有键找到有效路径
+            if not video_path:
+                for key, val in video.items():
+                    if isinstance(val, str) and os.path.exists(val):
+                        video_path = val
+                        break
        elif isinstance(video, str):
+            # 字符串格式：直接作为路径
            video_path = video
-        elif hasattr(video, "video"):
-            video_path = video.video
+        else:
+            # 对象格式：尝试常见属性
+            # 1. 尝试 __file 属性（VideoFromFile 对象）
+            if hasattr(video, "__file"):
+                video_path = video.__file
+            # 2. 尝试其他常见属性
+            elif hasattr(video, "video"):
+                video_path = video.video
+            elif hasattr(video, "path"):
+                video_path = video.path
+            elif hasattr(video, "filename"):
+                video_path = video.filename
+            # 3. 尝试从 __dict__ 中查找路径（支持私有属性如 _VideoFromFile__file）
+            elif hasattr(video, "__dict__"):
+                for attr_name, attr_value in video.__dict__.items():
+                    # 查找字符串类型的属性，且包含 file 或 path 关键字
+                    if isinstance(attr_value, str):
+                        if "file" in attr_name.lower() or "path" in attr_name.lower():
+                            # 验证路径是否有效
+                            if os.path.exists(attr_value):
+                                video_path = attr_value
+                                break
+                        # 如果属性值本身看起来像文件路径，也尝试使用
+                        elif os.path.exists(attr_value) and os.path.isfile(attr_value):
+                            video_path = attr_value
+                            break
        
        if not video_path or not os.path.exists(video_path):
            print(f"Google Gemini: 视频文件不存在或路径无效: {video_path}")
@@ -163,30 +439,70 @@ class GoogleGemini:
        
        mime_type = VIDEO_MIME_TYPES.get(ext, "video/mp4")
        
-        # 检查文件大小（限制 20MB）
-        file_size = os.path.getsize(video_path)
-        if file_size > 20 * 1024 * 1024:
-            raise ValueError(
-                f"视频文件过大 ({file_size / 1024 / 1024:.2f}MB)，"
-                f"请使用不超过 20MB 的视频文件"
-            )
-        
-        # 读取并编码视频
        try:
+            # 检查文件大小
+            file_size = os.path.getsize(video_path)
+            
+            # 如果超过 20MB，进行压缩
+            if file_size > MAX_FILE_SIZE:
+                video_path = self._compress_video(video_path)
+                temp_compressed_path = video_path
+                # 压缩后统一使用 mp4 格式
+                mime_type = "video/mp4"
+            
+            # 读取并编码视频
            with open(video_path, "rb") as f:
                video_bytes = f.read()
            
            b64_str = base64.b64encode(video_bytes).decode("utf-8")
            
+            # 清理临时文件
+            if temp_compressed_path and os.path.exists(temp_compressed_path):
+                try:
+                    os.remove(temp_compressed_path)
+                    print(f"Google Gemini: 临时压缩文件已清理")
+                except:
+                    pass
+            
            return {
                "mime_type": mime_type,
                "data": b64_str
            }
        
        except Exception as e:
-            print(f"Google Gemini: 读取视频文件失败 - {str(e)}")
+            # 清理临时文件
+            if temp_compressed_path and os.path.exists(temp_compressed_path):
+                try:
+                    os.remove(temp_compressed_path)
+                except:
+                    pass
+            
+            print(f"Google Gemini: 处理视频文件失败 - {str(e)}")
            return None
    
+    def _prepare_file_data(
+        self, 
+        file: Optional[FileData]
+    ) -> Optional[Dict[str, str]]:
+        """
+        准备文件数据
+        
+        从 FILE 类型提取文件数据
+        
+        Args:
+            file: FileData 对象（来自 LoadFile 节点）
+        
+        Returns:
+            文件数据字典，包含 mime_type 和 data
+        """
+        if file is None:
+            return None
+        
+        return {
+            "mime_type": file.mime_type,
+            "data": file.data
+        }
+    
    def _parse_dual_output(self, raw_response: Dict) -> Tuple[str, str]:
        """
        解析包含思考内容和主要内容的响应
@@ -214,16 +530,16 @@ class GoogleGemini:
                # 主要内容
                main_text = part.get("text", "")
        
-        return (main_text, thought_text)
+        return main_text
    
    def generate(
        self,
        模型: str,
        提示词: str,
-        思考深度: str,
-        系统指令: Optional[str] = None,
+        思考等级: str,
        图片: Optional[torch.Tensor] = None,
-        视频=None
+        视频=None,
+        文件: Optional[FileData] = None
    ) -> Tuple[str]:
        """
        生成文本
@@ -231,13 +547,13 @@ class GoogleGemini:
        Args:
            模型: 使用的模型名称
            提示词: 用户提示词
-            思考深度: 思考深度选项
-            系统指令: 系统级指令
+            思考等级: 思考等级选项
            图片: 输入图片
            视频: 输入视频
+            文件: 输入文件（PDF/TXT）
        
        Returns:
-            生成的文本 (STRING,)
+            (主要内容, 思考内容)
        """
        start_time = time.time()
        
@@ -259,6 +575,12 @@ class GoogleGemini:
            if video_data:
                print(f"Google Gemini: 输入视频 ({video_data['mime_type']})")
            
+            # 准备文件数据
+            document_data = self._prepare_file_data(文件)
+            if document_data:
+                file_type = "PDF" if document_data['mime_type'] == "application/pdf" else "TXT"
+                print(f"Google Gemini: 输入文件 ({file_type})")
+            
            # 构建输入描述
            input_desc = []
            if 提示词:
@@ -267,31 +589,32 @@ class GoogleGemini:
                input_desc.append(f"{len(image_data)}张图片")
            if video_data:
                input_desc.append("视频")
+            if document_data:
+                input_desc.append("文件")
            
            print(f"Google Gemini: 模型 = {模型}")
            print(f"Google Gemini: 多模态输入 ({', '.join(input_desc)})")
-            print(f"Google Gemini: 思考深度 = {思考深度}")
-            print(f"Google Gemini: 发送请求...")
+            print(f"Google Gemini: 思考等级 = {思考等级}")
            
            # 获取端点和构建请求体
-            endpoint = self.client.get_endpoint(model=模型, thinking_depth=思考深度)
+            endpoint = self.client.get_endpoint(model=模型)
            request_body = self.client.build_request_body(
                prompt=提示词,
-                system_instruction=系统指令,
+                model=模型,
+                thinking_level=思考等级,
                image_data=image_data,
-                video_data=video_data
+                video_data=video_data,
+                document_data=document_data
            )
            
-            # 根据是否有视频设置超时
-            timeout = 300 if video_data else 180
+            print(f"Google Gemini: 发送请求...")
            
            # 调用底层 API 获取原始响应
            async def get_raw_response():
                return await self.client.request_async(
                    endpoint, 
                    request_body, 
-                    session=None, 
-                    timeout=timeout
+                    session=None
                )
            
            # 在独立线程中执行异步请求
@@ -301,32 +624,56 @@ class GoogleGemini:
            elapsed = time.time() - start_time
            
            # 解析响应，分离主要内容和思考内容
-            main_text, thought_text = self._parse_dual_output(raw_response)
+            main_text = self._parse_dual_output(raw_response)
+            
+            # 打印响应 token 用量
+            usage = raw_response.get("usageMetadata", {})
+            prompt_tokens = usage.get("promptTokenCount", 0)
+            candidates_tokens = usage.get("candidatesTokenCount", 0)
+            thoughts_tokens = usage.get("thoughtsTokenCount", 0)
+            total_tokens = usage.get("totalTokenCount", 0)
+            finish_reason = ""
+            candidates = raw_response.get("candidates", [])
+            if candidates:
+                finish_reason = candidates[0].get("finishReason", "")
            
-            # 输出信息
            print(f"Google Gemini: 生成完成 (耗时: {elapsed:.2f}s)")
+            print(f"Google Gemini: finishReason = {finish_reason}")
+            print(f"Google Gemini: Token 用量 — 输入: {prompt_tokens}, 输出: {candidates_tokens}, 思考: {thoughts_tokens}, 合计: {total_tokens}")
            print(f"Google Gemini: 主要内容长度: {len(main_text)} 字符")
-            print(f"Google Gemini: 思考内容长度: {len(thought_text)} 字符")
            
            # 输出预览
            if main_text:
                preview = main_text[:100] + "..." if len(main_text) > 100 else main_text
                print(f"Google Gemini: 主要内容预览: {preview}")
            
-            return (main_text, thought_text)
-        
-        except ValueError as e:
-            # 检测是否为授权错误
+            return (main_text,)
            if str(e) == "未授权！":
                print("请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
            else:
-                print(f"Google Gemini: 输入错误 - {str(e)}")
-            raise
+                # 用户输入错误 - 只显示简洁信息
+                error_msg = str(e).split('\n')[0]  # 只取第一行
+                print(f"Google Gemini: ❌ {error_msg}")
+                raise ValueError(error_msg) from None
        
        except RuntimeError as e:
-            print(f"Google Gemini: API 错误 - {str(e)}")
-            raise
+            # 日志只打第一行；报错框展示完整多行
+            error_full = str(e)
+            print(f"Google Gemini: ❌ {error_full.split('\n')[0]}")
+            raise RuntimeError(error_full) from None
        
        except Exception as e:
-            print(f"Google Gemini: 未知错误 - {str(e)}")
-            raise
+            # 其他未知错误 - 只显示简洁信息
+            error_msg = str(e).split('\n')[0]
+            print(f"Google Gemini: ❌ {error_msg}")
+            raise type(e)(error_msg) from None
+        
+        finally:
+            if self.client is not None:
+                try:
+                    balance_data = self.client.query_balance_sync()
+                    balance_info = self.client.format_balance_info(balance_data)
+                    print(f"Google Gemini: {balance_info}")
+                except Exception:
+                    pass
@@ -0,0 +1,824 @@
+"""
+o1key GPT Image 节点
+支持 gpt-image-1 / gpt-image-1.5 模型的文生图、图生图、图像编辑（带蒙版）
+"""
+
+import os
+import time
+from typing import List, Optional, Tuple
+
+from PIL import Image
+
+from ..clients.gpt_image_client import GptImageClient
+from ..utils.image_utils import parse_batch_prompts, pil_to_tensor, tensor_to_pil
+from ..utils.config import NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.file_utils import (
+    ImageInfo,
+    generate_timestamp_filename,
+    load_images_from_folder,
+    pair_images_by_name,
+    pair_images_cartesian,
+    save_image,
+)
+
+try:
+    from comfy.model_management import processing_interrupted, InterruptProcessingException
+    _INTERRUPT_AVAILABLE = True
+except ImportError:
+    _INTERRUPT_AVAILABLE = False
+    processing_interrupted = lambda: False
+    InterruptProcessingException = RuntimeError
+
+try:
+    from comfy.utils import ProgressBar
+    _PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    _PROGRESS_BAR_AVAILABLE = False
+
+try:
+    import folder_paths
+    _FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    _FOLDER_PATHS_AVAILABLE = False
+
+
+def _make_node_progress_callback(progress_bar, task_index: int, total_tasks: int):
+    if progress_bar is None:
+        return None
+
+    total_units = max(1, total_tasks) * 100
+    base_units = max(0, task_index - 1) * 100
+    last_pct = {"value": -1}
+
+    def _callback(pct: int):
+        try:
+            pct_value = int(round(float(pct)))
+        except (TypeError, ValueError):
+            return
+        pct_value = max(0, min(100, pct_value))
+        if pct_value < last_pct["value"]:
+            return
+        last_pct["value"] = pct_value
+        progress_bar.update_absolute(
+            min(total_units, base_units + pct_value),
+            total_units,
+        )
+
+    return _callback
+
+
+def _resolve_async_size(value: str) -> str:
+    value = (value or "").strip()
+    if not value or value == "智能" or value.lower() == "auto":
+        return "auto"
+
+    first_part = value.split("（")[0].strip()
+    normalized_size = first_part.lower().replace("*", "x").replace("×", "x")
+    size_parts = [part.strip() for part in normalized_size.split("x")]
+    if len(size_parts) == 2 and all(part.isdigit() for part in size_parts):
+        return f"{int(size_parts[0])}x{int(size_parts[1])}"
+
+    allowed = {"auto", "1024x1024", "1K", "2K", "4K"}
+    if first_part in allowed:
+        return first_part
+
+    if "4K" in value:
+        return "4K"
+    if "2K" in value:
+        return "2K"
+    if "1K" in value:
+        return "1K"
+
+    return "auto"
+
+
+class O1keyGPTImage:
+    """
+    o1key GPT Image 节点
+
+    功能：
+      - 文生图：仅提供 prompt
+      - 图生图：提供 prompt + 图片（无遮罩）
+      - 图像编辑：提供 prompt + 图片 + 遮罩（白色区域将被替换）
+      - 批量模式：prompt 中用单独一行 --- 分隔多条提示词
+
+    参数：
+      - prompt   : 文本提示词（多行；用 --- 独占一行分隔批量提示词）
+      - 模型     : 模型选择
+      - 分辨率   : 图像尺寸（auto 让 API 自动决定）
+      - 生图数量 : 每条提示词生成数量 1-8
+      - 质量     : 生成质量
+      - seed     : 随机种子（0 表示不指定）
+      - 图片     : 可选参考图（用于图生图或编辑）
+      - 遮罩     : 可选蒙版（白色区域将被替换）
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        # 创建9个独立的参考图输入
+        optional_inputs = {}
+        for i in range(1, 10):
+            optional_inputs[f"参考图{i}"] = ("IMAGE", {
+                "tooltip": f"Optional reference image {i} for image editing.",
+            })
+
+        optional_inputs["模型"] = ([
+            "gpt-image-2-按量",
+            "gpt-image-2-次卡",
+        ], {
+            "default": "gpt-image-2-次卡",
+        })
+        optional_inputs["网络"] = (NETWORK_ROUTE_OPTIONS, {
+            "default": "全球加速",
+        })
+        optional_inputs["分辨率"] = ([
+            "智能",
+            # ── 1K ──
+            "1024x1024（1K 正方形 1:1）",
+            "1536x1024（1K 横版 3:2）",
+            "1024x1536（1K 竖版 2:3）",
+            "1360x1024（1K 横版 4:3）",
+            "1024x1360（1K 竖版 3:4）",
+            "1824x1024（1K 横版 16:9）",
+            "1024x1824（1K 竖版 9:16）",
+            # ── 2K ──
+            "2048x2048（2K 正方形 1:1）",
+            "3072x2048（2K 横版 3:2）",
+            "2048x3072（2K 竖版 2:3）",
+            "2736x2048（2K 横版 4:3）",
+            "2048x2736（2K 竖版 3:4）",
+            "3648x2048（2K 横版 16:9）",
+            "2048x3648（2K 竖版 9:16）",
+            # ── 4K ──
+            "2880x2880（4K 正方形 1:1）",
+            "3504x2336（4K 横版 3:2）",
+            "2336x3504（4K 竖版 2:3）",
+            "3264x2448（4K 横版 4:3）",
+            "2448x3264（4K 竖版 3:4）",
+            "3840x2160（4K 横版 16:9）",
+            "2160x3840（4K 竖版 9:16）",
+        ], {
+            "default": "智能",
+            "tooltip": "Image size (智能 = API decides)",
+        })
+        optional_inputs["生图数量"] = ("INT", {
+            "default": 1,
+            "min": 1,
+            "max": 8,
+            "step": 1,
+            "display": "number",
+            "tooltip": "How many images to generate per prompt",
+        })
+        optional_inputs["质量"] = (["高", "中", "低", "自动"], {
+            "default": "自动",
+            "tooltip": "Image quality: 高=high, 中=medium, 低=low, 自动=auto",
+        })
+        optional_inputs["输出格式"] = (["png", "jpeg", "webp"], {
+            "default": "jpeg",
+            "tooltip": "Generated image output format",
+        })
+        optional_inputs["seed"] = ("INT", {
+            "default": 0,
+            "min": 0,
+            "max": 2**31 - 1,
+            "step": 1,
+            "display": "number",
+            "control_after_generate": True,
+            "tooltip": "Random seed (0 = not specified)",
+        })
+        optional_inputs["遮罩"] = ("MASK", {
+            "tooltip": "Optional mask for inpainting (white areas will be replaced)",
+        })
+
+        return {
+            "required": {
+                "prompt": ("STRING", {
+                    "default": "",
+                    "multiline": True,
+                    "tooltip": "Text prompt for GPT Image. Use --- on its own line to separate batch prompts.",
+                }),
+            },
+            "optional": optional_inputs,
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("IMAGE",)
+    FUNCTION = "generate"
+    CATEGORY = "o1key/image"
+    OUTPUT_NODE = False
+
+    def generate(
+        self,
+        prompt: str,
+        模型: str = "gpt-image-2-次卡",
+        网络: str = "全球加速",
+        分辨率: str = "auto",
+        质量: str = "自动",
+        输出格式: str = "jpeg",
+        生图数量: int = 1,
+        seed: int = 0,
+        遮罩=None,
+        **kwargs,
+    ):
+        """
+        生成图像（文生图 / 图生图 / 图像编辑 / 批量提示词）
+
+        路由逻辑：
+          - 无图片           → generations 接口（文生图）
+          - 有图片，无遮罩   → edits 接口（图生图）
+          - 有图片，有遮罩   → edits 接口（图像编辑 + 蒙版）
+          - prompt 含 ---    → 批量模式，逐条调用上述接口
+        """
+        start_time = time.time()
+
+        # ── 0. 收集多参考图输入 ────────────────────────────────────────────────
+        reference_tensors = []
+        for i in range(1, 10):
+            key = f"参考图{i}"
+            if key in kwargs and kwargs[key] is not None:
+                reference_tensors.append(kwargs[key])
+
+        图片 = reference_tensors if reference_tensors else None
+
+        # ── 1. 参数校验 ───────────────────────────────────────────────────────
+        if 遮罩 is not None and 图片 is None:
+            raise ValueError("提供了遮罩但未提供图片，请同时提供图片和遮罩")
+
+        # ── 2. 解析分辨率显示值 → API 参数值 ──────────────────────────────────
+        size = _resolve_async_size(分辨率)
+
+        # ── 2b. 解析模型显示值 → API 参数值 ───────────────────────────────────
+        _model_map = {"gpt-image-2-次卡": "gpt-image-2-c", "gpt-image-2-按量": "gpt-image-2"}
+        model = _model_map.get(模型, 模型)
+
+        # ── 2c. 解析质量显示值 → API 参数值 ───────────────────────────────────
+        _quality_map = {"高": "high", "中": "medium", "低": "low", "自动": "auto"}
+        quality = _quality_map.get(质量, "auto")
+
+        # ── 3. 创建客户端 ─────────────────────────────────────────────────────
+        try:
+            client = GptImageClient()
+            client.base_url = get_base_url_by_route(网络)
+        except ValueError as e:
+            if str(e) == "未授权！":
+                print("[o1key GPT Image] 请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            raise
+
+        try:
+            # ── 4. 解析批量提示词 ─────────────────────────────────────────────
+            batch_prompts = parse_batch_prompts(prompt)
+
+            # ── 5. 调用 API ───────────────────────────────────────────────────
+            all_pil_images = []
+            progress_total = len(batch_prompts) if batch_prompts else 1
+            progress_bar = ProgressBar(progress_total * 100) if _PROGRESS_BAR_AVAILABLE else None
+
+            if batch_prompts:
+                # 批量模式：逐条提示词调用
+                total = len(batch_prompts)
+                print(f"[o1key GPT Image] 批量模式 | {total} 条提示词 | 每条生成 {生图数量} 张")
+                for idx, p in enumerate(batch_prompts, 1):
+                    if _INTERRUPT_AVAILABLE and processing_interrupted():
+                        print("[o1key GPT Image] 用户取消，已中断批量生成")
+                        raise InterruptProcessingException()
+                    try:
+                        pil_images = client.generate_image_async_sync(
+                            prompt=p,
+                            model=model,
+                            quality=quality,
+                            size=size,
+                            n=生图数量,
+                            seed=seed,
+                            image_tensor=图片,
+                            mask_tensor=遮罩,
+                            output_format=输出格式,
+                            progress_callback=_make_node_progress_callback(progress_bar, idx, total),
+                        )
+                        all_pil_images.extend(pil_images)
+                        snippet = p[:30] + ("..." if len(p) >= 30 else "")
+                        print(f"[o1key GPT Image] [{idx}/{total}] ✓ {snippet}")
+                    except InterruptProcessingException:
+                        raise
+                    except Exception as e:
+                        error_msg = str(e).split('\n')[0]
+                        snippet = p[:30] + ("..." if len(p) >= 30 else "")
+                        print(f"[o1key GPT Image] [{idx}/{total}] ❌ {snippet} → {error_msg}")
+                    if progress_bar is not None:
+                        progress_bar.update_absolute(idx * 100, total * 100)
+            else:
+                # 单提示词模式
+                if not prompt or not prompt.strip():
+                    raise ValueError("提示词不能为空")
+                try:
+                    pil_images = client.generate_image_async_sync(
+                        prompt=prompt,
+                        model=model,
+                        quality=quality,
+                        size=size,
+                        n=生图数量,
+                        seed=seed,
+                        image_tensor=图片,
+                        mask_tensor=遮罩,
+                        output_format=输出格式,
+                        progress_callback=_make_node_progress_callback(progress_bar, 1, 1),
+                    )
+                    all_pil_images.extend(pil_images)
+                except InterruptProcessingException:
+                    raise
+                except Exception as e:
+                    error_msg = str(e).split('\n')[0]
+                    print(f"[o1key GPT Image] ❌ {error_msg}")
+                    raise RuntimeError(error_msg) from None
+
+            # ── 6. 检查是否有可用图像 ─────────────────────────────────────────
+            if not all_pil_images:
+                raise RuntimeError("所有提示词均生成失败，无可用图像输出")
+
+            # ── 7. PIL → tensor ───────────────────────────────────────────────
+            output_tensor = GptImageClient._pil_list_to_tensor(all_pil_images)
+
+            # ── 8. 完成日志 ───────────────────────────────────────────────────
+            elapsed = time.time() - start_time
+            print(
+                f"[o1key GPT Image] 完成！耗时 {elapsed:.1f}s，"
+                f"输出 {output_tensor.shape[0]} 张 "
+                f"{output_tensor.shape[2]}×{output_tensor.shape[1]}"
+            )
+
+            return (output_tensor,)
+
+        finally:
+            self._print_balance(client)
+
+    def _print_balance(self, client):
+        try:
+            balance_data = client.query_balance_sync()
+            balance_info = client.format_balance_info(balance_data)
+            print(f"[o1key GPT Image] {balance_info}")
+        except Exception:
+            pass
+
+
+class O1keyGPTImageBatch:
+    """
+    o1key GPT Image 批量节点
+
+    复用 BatchNanoBananaPro 的批量思路：
+      - 从文件夹批量加载图片
+      - 按文件名同名 / 1*N / 不配对 三种模式创建任务
+      - 可追加节点手动输入参考图
+      - prompt 支持用独占一行 --- 展开为多提示词任务
+      - 每个任务调用 GPT Image 客户端并保存到磁盘
+    """
+
+    PAIRING_MODES = ["按相同图片命名", "1*N", "不配对"]
+    IMAGE_FORMATS = ["原始", "JPEG", "PNG", "WebP"]
+    MODEL_OPTIONS = ["gpt-image-2-按量", "gpt-image-2-次卡"]
+    QUALITY_OPTIONS = ["高", "中", "低", "自动"]
+    RESOLUTION_OPTIONS = [
+        "智能",
+        "1024x1024（1K 正方形 1:1）",
+        "1536x1024（1K 横版 3:2）",
+        "1024x1536（1K 竖版 2:3）",
+        "1360x1024（1K 横版 4:3）",
+        "1024x1360（1K 竖版 3:4）",
+        "1824x1024（1K 横版 16:9）",
+        "1024x1824（1K 竖版 9:16）",
+        "2048x2048（2K 正方形 1:1）",
+        "3072x2048（2K 横版 3:2）",
+        "2048x3072（2K 竖版 2:3）",
+        "2736x2048（2K 横版 4:3）",
+        "2048x2736（2K 竖版 3:4）",
+        "3648x2048（2K 横版 16:9）",
+        "2048x3648（2K 竖版 9:16）",
+        "2880x2880（4K 正方形 1:1）",
+        "3504x2336（4K 横版 3:2）",
+        "2336x3504（4K 竖版 2:3）",
+        "3264x2448（4K 横版 4:3）",
+        "2448x3264（4K 竖版 3:4）",
+        "3840x2160（4K 横版 16:9）",
+        "2160x3840（4K 竖版 9:16）",
+    ]
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        optional_inputs = {}
+        for image_index in range(1, 10):
+            optional_inputs[f"参考图{image_index}"] = ("IMAGE", {
+                "tooltip": "追加到每个批量任务末尾的固定参考图。",
+            })
+
+        optional_inputs["遮罩"] = ("MASK", {
+            "tooltip": "可选蒙版，会应用到每个任务的第一张参考图；请确保尺寸一致。",
+        })
+        optional_inputs["图片配对模式"] = (cls.PAIRING_MODES, {
+            "default": "不配对",
+            "tooltip": "文件夹图片的组合方式；手动参考图只追加，不参与配对。",
+        })
+
+        return {
+            "required": {
+                "prompt": ("STRING", {
+                    "default": "",
+                    "multiline": True,
+                    "tooltip": "提示词；可用独占一行的 --- 分隔多条批量提示词。",
+                }),
+                "模型": (cls.MODEL_OPTIONS, {
+                    "default": "gpt-image-2-次卡",
+                }),
+                "网络": (NETWORK_ROUTE_OPTIONS, {
+                    "default": "全球加速",
+                }),
+                "分辨率": (cls.RESOLUTION_OPTIONS, {
+                    "default": "智能",
+                }),
+                "生图数量": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 8,
+                    "step": 1,
+                    "display": "number",
+                }),
+                "质量": (cls.QUALITY_OPTIONS, {
+                    "default": "自动",
+                }),
+                "seed": ("INT", {
+                    "default": 0,
+                    "min": 0,
+                    "max": 2**31 - 1,
+                    "step": 1,
+                    "display": "number",
+                    "control_after_generate": True,
+                }),
+                "图片格式": (cls.IMAGE_FORMATS, {
+                    "default": "原始",
+                }),
+                "文件夹1": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                }),
+                "文件夹2": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                }),
+                "文件夹3": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                }),
+                "文件夹4": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                }),
+                "文件夹5": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                }),
+                "保存路径": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                    "tooltip": "为空时优先使用 ComfyUI 默认 output 目录。",
+                }),
+            },
+            "optional": optional_inputs,
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("IMAGE",)
+    FUNCTION = "process_batch"
+    CATEGORY = "o1key/image"
+    OUTPUT_NODE = False
+
+    def _load_folders(self, folders: List[str]) -> List[List[ImageInfo]]:
+        image_lists = []
+        for folder_index, folder in enumerate(folders, 1):
+            if not folder or not folder.strip():
+                continue
+            try:
+                loaded_images = load_images_from_folder(folder)
+                if loaded_images:
+                    image_lists.append(loaded_images)
+            except ValueError as error:
+                print(f"[o1key GPT Image Batch] 文件夹{folder_index} 加载失败 - {error}")
+        return image_lists
+
+    def _create_pairs(
+        self,
+        image_lists: List[List[ImageInfo]],
+        pairing_mode: str,
+        manual_images: Optional[List[ImageInfo]] = None,
+    ) -> List[Tuple[ImageInfo, ...]]:
+        if pairing_mode == "不配对":
+            if len(image_lists) > 1:
+                raise ValueError("「不配对」模式只支持单个文件夹，请清空其他文件夹路径")
+
+            if image_lists and manual_images:
+                return [
+                    (folder_image,) + tuple(manual_images)
+                    for folder_image in image_lists[0]
+                ]
+            if image_lists:
+                return [(folder_image,) for folder_image in image_lists[0]]
+            return []
+
+        if not image_lists:
+            return []
+
+        if len(image_lists) == 1:
+            base_pairs = [(folder_image,) for folder_image in image_lists[0]]
+        elif pairing_mode == "按相同图片命名":
+            base_pairs = list(pair_images_by_name(*image_lists))
+        else:
+            base_pairs = list(pair_images_cartesian(*image_lists))
+
+        if manual_images:
+            manual_tuple = tuple(manual_images)
+            base_pairs = [pair + manual_tuple for pair in base_pairs]
+
+        return base_pairs
+
+    def _collect_manual_images(self, kwargs) -> List[ImageInfo]:
+        manual_images = []
+        for image_index in range(1, 10):
+            key = f"参考图{image_index}"
+            if key not in kwargs or kwargs[key] is None:
+                continue
+            for tensor_index, image in enumerate(tensor_to_pil(kwargs[key])):
+                manual_images.append(ImageInfo(
+                    image=image,
+                    filename=f"manual_{image_index}_{tensor_index}",
+                    extension=".png",
+                    source_path="",
+                ))
+        return manual_images
+
+    @staticmethod
+    def _pair_to_tensors(pair: Tuple[ImageInfo, ...]) -> List:
+        return [pil_to_tensor([image_info.image]) for image_info in pair]
+
+    @staticmethod
+    def _resolve_size(分辨率: str) -> str:
+        return _resolve_async_size(分辨率)
+
+    @staticmethod
+    def _resolve_model(模型: str) -> str:
+        model_map = {
+            "gpt-image-2-次卡": "gpt-image-2-c",
+            "gpt-image-2-按量": "gpt-image-2",
+        }
+        return model_map.get(模型, 模型)
+
+    @staticmethod
+    def _resolve_quality(质量: str) -> str:
+        quality_map = {"高": "high", "中": "medium", "低": "low", "自动": "auto"}
+        return quality_map.get(质量, "auto")
+
+    @staticmethod
+    def _resolve_output_format(图片格式: str) -> str:
+        output_format_map = {
+            "JPEG": "jpeg",
+            "PNG": "png",
+            "WebP": "webp",
+        }
+        return output_format_map.get(图片格式, "png")
+
+    @staticmethod
+    def _ensure_output_folder(保存路径: str) -> str:
+        output_folder = (保存路径 or "").strip()
+        if not output_folder and _FOLDER_PATHS_AVAILABLE:
+            output_folder = folder_paths.get_output_directory()
+            print(f"[o1key GPT Image Batch] 未设置保存路径，使用 ComfyUI 默认 output 目录: {output_folder}")
+
+        if not output_folder:
+            raise ValueError("未设置保存路径，且当前环境无法获取 ComfyUI 默认 output 目录")
+
+        os.makedirs(output_folder, exist_ok=True)
+        test_path = os.path.join(output_folder, ".write_test")
+        with open(test_path, "w", encoding="utf-8") as test_file:
+            test_file.write("test")
+        os.remove(test_path)
+        return output_folder
+
+    @staticmethod
+    def _save_images(
+        images: List[Image.Image],
+        output_folder: str,
+        image_format: str,
+        base_filename: Optional[str] = None,
+    ) -> List[str]:
+        format_ext_map = {"JPEG": ".jpg", "PNG": ".png", "WebP": ".webp"}
+        save_ext = format_ext_map.get(image_format, ".png")
+        saved_files = []
+
+        for image in images:
+            if base_filename:
+                counter = 0
+                while True:
+                    suffix = "" if counter == 0 else f"+{counter}"
+                    filename = f"{base_filename}{suffix}{save_ext}"
+                    output_path = os.path.join(output_folder, filename)
+                    if not os.path.exists(output_path):
+                        break
+                    counter += 1
+            else:
+                output_path = generate_timestamp_filename(
+                    output_folder=output_folder,
+                    extension=save_ext,
+                )
+
+            if image_format == "JPEG":
+                if image.mode != "RGB":
+                    image = image.convert("RGB")
+                image.save(output_path, quality=100)
+            elif image_format == "WebP":
+                image.save(output_path, lossless=True)
+            else:
+                save_image(image, output_path)
+
+            saved_files.append(output_path)
+
+        return saved_files
+
+    def process_batch(
+        self,
+        prompt: str,
+        模型: str,
+        网络: str,
+        分辨率: str,
+        生图数量: int,
+        质量: str,
+        seed: int,
+        图片格式: str,
+        文件夹1: str,
+        文件夹2: str,
+        文件夹3: str,
+        文件夹4: str,
+        文件夹5: str,
+        保存路径: str = "",
+        图片配对模式: str = "不配对",
+        遮罩=None,
+        **kwargs,
+    ):
+        start_time = time.time()
+        client = None
+
+        try:
+            if not prompt or not prompt.strip():
+                raise ValueError("提示词不能为空")
+
+            folders = [文件夹1, 文件夹2, 文件夹3, 文件夹4, 文件夹5]
+            if not any(folder and folder.strip() for folder in folders):
+                raise ValueError("请至少填写一个文件夹路径，该节点专为批量文件夹处理设计")
+
+            image_lists = self._load_folders(folders)
+            total_folder_images = sum(len(image_list) for image_list in image_lists)
+            if total_folder_images == 0:
+                raise ValueError("文件夹中未找到任何图片，请检查文件夹路径是否正确")
+
+            manual_images = self._collect_manual_images(kwargs)
+            pairs = self._create_pairs(
+                image_lists=image_lists,
+                pairing_mode=图片配对模式,
+                manual_images=manual_images if manual_images else None,
+            )
+            if not pairs:
+                raise ValueError("配对结果为空，请检查输入")
+
+            batch_prompts = parse_batch_prompts(prompt)
+            prompts_per_task = None
+            if batch_prompts:
+                expanded_pairs = []
+                expanded_prompts = []
+                for pair in pairs:
+                    for batch_prompt in batch_prompts:
+                        expanded_pairs.append(pair)
+                        expanded_prompts.append(batch_prompt)
+                pairs = expanded_pairs
+                prompts_per_task = expanded_prompts
+
+            total_tasks = len(pairs)
+            if batch_prompts:
+                print(
+                    f"[o1key GPT Image Batch] 批量任务 | {图片配对模式} × "
+                    f"{len(batch_prompts)} 个提示词 | 共 {total_tasks} 任务"
+                )
+            else:
+                print(f"[o1key GPT Image Batch] 批量任务 | {图片配对模式} | 共 {total_tasks} 任务")
+
+            output_folder = self._ensure_output_folder(保存路径)
+            size = self._resolve_size(分辨率)
+            model = self._resolve_model(模型)
+            quality = self._resolve_quality(质量)
+            output_format = self._resolve_output_format(图片格式)
+
+            client = GptImageClient()
+            client.base_url = get_base_url_by_route(网络)
+
+            progress_bar = ProgressBar(total_tasks * 100) if _PROGRESS_BAR_AVAILABLE else None
+            results = []
+            all_saved_files = []
+
+            for task_index, pair in enumerate(pairs, 1):
+                if _INTERRUPT_AVAILABLE and processing_interrupted():
+                    print("[o1key GPT Image Batch] 用户取消，已中断批量生成")
+                    raise InterruptProcessingException()
+
+                task_prompt = prompts_per_task[task_index - 1] if prompts_per_task else prompt
+                base_filename = pair[0].filename if pair else None
+                result = {
+                    "task_index": task_index,
+                    "success": False,
+                    "generated_count": 0,
+                    "saved_files": [],
+                    "error": None,
+                }
+
+                try:
+                    pil_images = client.generate_image_async_sync(
+                        prompt=task_prompt,
+                        model=model,
+                        quality=quality,
+                        size=size,
+                        n=生图数量,
+                        seed=seed,
+                        image_tensor=self._pair_to_tensors(pair),
+                        mask_tensor=遮罩,
+                        output_format=output_format,
+                        progress_callback=_make_node_progress_callback(progress_bar, task_index, total_tasks),
+                    )
+                    saved_files = self._save_images(
+                        images=pil_images,
+                        output_folder=output_folder,
+                        image_format=图片格式,
+                        base_filename=base_filename,
+                    )
+                    result["success"] = bool(pil_images)
+                    result["generated_count"] = len(pil_images)
+                    result["saved_files"] = saved_files
+                    all_saved_files.extend(saved_files)
+                    print(f"[o1key GPT Image Batch] [{task_index}/{total_tasks}] ✓ {base_filename or 'task'}")
+                except InterruptProcessingException:
+                    raise
+                except Exception as error:
+                    error_msg = str(error).split("\n")[0]
+                    result["error"] = error_msg
+                    print(f"[o1key GPT Image Batch] [{task_index}/{total_tasks}] ❌ {base_filename or 'task'} → {error_msg}")
+
+                results.append(result)
+                if progress_bar is not None:
+                    progress_bar.update_absolute(task_index * 100, total_tasks * 100)
+
+            success_count = sum(1 for result in results if result.get("success", False))
+            total_generated = sum(result.get("generated_count", 0) for result in results)
+            if success_count == 0:
+                raise RuntimeError("所有批量任务均生成失败，无可用图像输出")
+
+            output_images = []
+            for file_path in all_saved_files[-10:]:
+                try:
+                    loaded_image = Image.open(file_path)
+                    loaded_image.load()
+                    output_images.append(loaded_image)
+                except Exception as error:
+                    print(f"[o1key GPT Image Batch] 无法加载输出图片 {file_path} - {error}")
+
+            if not output_images:
+                output_images = [Image.new("RGBA", (512, 512), (128, 128, 128, 255))]
+
+            output_tensor = GptImageClient._pil_list_to_tensor(output_images)
+            elapsed = time.time() - start_time
+            print("=" * 60)
+            print(
+                f"[o1key GPT Image Batch] 完成！耗时 {elapsed:.1f}s | "
+                f"成功 {success_count}/{total_tasks} | 生成 {total_generated} 张"
+            )
+            print(f"[o1key GPT Image Batch] 保存路径: {output_folder}")
+            if all_saved_files:
+                print(f"[o1key GPT Image Batch] 最新保存文件: {all_saved_files[-1]}")
+
+            failed_results = [result for result in results if not result.get("success", False)]
+            if failed_results:
+                print(f"[o1key GPT Image Batch] 失败任务: {len(failed_results)} 个")
+                for failed_result in failed_results[:3]:
+                    print(
+                        f"  - #{failed_result.get('task_index')}: "
+                        f"{failed_result.get('error', '未知错误')}"
+                    )
+
+            return (output_tensor,)
+
+        except ValueError as error:
+            if str(error) == "未授权！":
+                print("[o1key GPT Image Batch] 请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            raise ValueError(str(error)) from None
+        except RuntimeError as error:
+            raise RuntimeError(str(error)) from None
+        finally:
+            if client is not None:
+                try:
+                    balance_data = client.query_balance_sync()
+                    balance_info = client.format_balance_info(balance_data)
+                    print(f"[o1key GPT Image Batch] {balance_info}")
+                except Exception:
+                    pass
@@ -0,0 +1,398 @@
+"""
+Merged grid image splitter.
+
+This node is designed for AI-generated contact sheets such as 3x3 or 2x3
+grids. Auto mode scores common layouts by looking for strong seams or flat
+separator bands near the expected grid lines, then crops each cell.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import List, Sequence, Tuple
+
+import numpy as np
+import torch
+from PIL import Image
+
+from ..utils.image_utils import pil_to_tensor, tensor_to_pil
+
+
+_AUTO_LAYOUTS: Sequence[Tuple[int, int]] = (
+    (3, 3),
+    (2, 3),
+    (3, 2),
+    (2, 2),
+    (1, 2),
+    (2, 1),
+    (1, 3),
+    (3, 1),
+    (4, 4),
+    (3, 4),
+    (4, 3),
+)
+
+_LAYOUTS = [
+    "auto",
+    "1x2",
+    "2x1",
+    "1x3",
+    "3x1",
+    "2x2",
+    "2x3",
+    "3x2",
+    "3x3",
+    "3x4",
+    "4x3",
+    "4x4",
+    "custom",
+]
+
+
+@dataclass(frozen=True)
+class _AxisCut:
+    seam: int
+    span_start: int
+    span_end: int
+    score: float
+
+
+@dataclass(frozen=True)
+class _AxisPlan:
+    intervals: List[Tuple[int, int]]
+    cuts: List[_AxisCut]
+    score: float
+
+
+def _to_float_array(image: Image.Image) -> np.ndarray:
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+    return np.asarray(image).astype(np.float32) / 255.0
+
+
+def _axis_texture(arr: np.ndarray, axis: str) -> np.ndarray:
+    if axis == "x":
+        profile = arr.std(axis=(0, 2))
+    else:
+        profile = arr.std(axis=(1, 2))
+    high = np.percentile(profile, 95) + 1e-6
+    return np.clip(profile / high, 0.0, 1.0)
+
+
+def _axis_edge(arr: np.ndarray, axis: str) -> np.ndarray:
+    if axis == "x":
+        diff = np.abs(np.diff(arr, axis=1)).mean(axis=(0, 2))
+        length = arr.shape[1]
+    else:
+        diff = np.abs(np.diff(arr, axis=0)).mean(axis=(1, 2))
+        length = arr.shape[0]
+
+    padded = np.zeros(length, dtype=np.float32)
+    if diff.size:
+        padded[1:] = diff
+    high = np.percentile(padded, 95) + 1e-6
+    return np.clip(padded / high, 0.0, 1.5)
+
+
+def _smooth(profile: np.ndarray, radius: int = 2) -> np.ndarray:
+    if radius <= 0 or profile.size < radius * 2 + 1:
+        return profile
+    kernel = np.ones(radius * 2 + 1, dtype=np.float32) / float(radius * 2 + 1)
+    return np.convolve(profile, kernel, mode="same")
+
+
+def _separator_span(
+    texture: np.ndarray,
+    seam: int,
+    search_px: int,
+    min_separator_px: int,
+) -> Tuple[int, int]:
+    length = texture.size
+    if length <= 1:
+        return 0, length
+
+    limit = max(1, min(search_px, length // 8))
+    threshold = max(0.08, min(0.28, float(np.percentile(texture, 12)) * 1.8))
+
+    left = seam
+    while left > 0 and seam - left < limit and texture[left - 1] <= threshold:
+        left -= 1
+
+    right = seam
+    while right < length and right - seam < limit and texture[right] <= threshold:
+        right += 1
+
+    if right - left >= max(1, min_separator_px):
+        return left, right
+
+    return seam, seam
+
+
+def _edge_trim(texture: np.ndarray, search_px: int, min_cell: int) -> Tuple[int, int]:
+    length = texture.size
+    if length <= 2:
+        return 0, length
+
+    max_trim = max(0, min(search_px * 2, min_cell // 3, length // 6))
+    if max_trim <= 0:
+        return 0, length
+
+    threshold = max(0.08, min(0.24, float(np.percentile(texture, 12)) * 1.6))
+
+    start = 0
+    while start < max_trim and texture[start] <= threshold:
+        start += 1
+
+    end = length
+    while length - end < max_trim and end > start + min_cell and texture[end - 1] <= threshold:
+        end -= 1
+
+    return start, end
+
+
+def _axis_plan(
+    arr: np.ndarray,
+    cells: int,
+    axis: str,
+    search_px: int,
+    crop_separators: bool,
+    trim_outer: bool,
+    min_separator_px: int,
+) -> _AxisPlan:
+    length = arr.shape[1] if axis == "x" else arr.shape[0]
+    if cells <= 1:
+        return _AxisPlan(intervals=[(0, length)], cuts=[], score=0.0)
+
+    raw_texture = _axis_texture(arr, axis)
+    raw_edge = _axis_edge(arr, axis)
+    texture = _smooth(raw_texture, radius=2)
+    edge = _smooth(raw_edge, radius=1)
+    evidence = np.maximum(edge, (1.0 - texture) * 0.75)
+    exact_evidence = np.maximum(raw_edge, (1.0 - raw_texture) * 0.75)
+
+    cuts: List[_AxisCut] = []
+    scores: List[float] = []
+    for idx in range(1, cells):
+        expected = round(length * idx / cells)
+        start = max(1, expected - search_px)
+        end = min(length - 1, expected + search_px)
+        if start >= end:
+            seam = expected
+            score = 0.0
+        else:
+            window = evidence[start:end + 1]
+            offset = int(window.argmax())
+            coarse = start + offset
+            fine_start = max(start, coarse - 2)
+            fine_end = min(end, coarse + 2)
+            fine_window = exact_evidence[fine_start:fine_end + 1]
+            seam = fine_start + int(fine_window.argmax())
+            score = float(window[offset])
+
+        span_start, span_end = _separator_span(
+            raw_texture,
+            seam,
+            search_px=search_px,
+            min_separator_px=min_separator_px,
+        )
+        cuts.append(_AxisCut(seam=seam, span_start=span_start, span_end=span_end, score=score))
+        scores.append(score)
+
+    min_cell = max(1, length // cells)
+    outer_start, outer_end = _edge_trim(raw_texture, search_px, min_cell) if trim_outer else (0, length)
+
+    intervals: List[Tuple[int, int]] = []
+    cursor = outer_start
+    for cut in cuts:
+        split_start = cut.span_start if crop_separators else cut.seam
+        split_end = cut.span_end if crop_separators else cut.seam
+        intervals.append((cursor, split_start))
+        cursor = split_end
+    intervals.append((cursor, outer_end))
+
+    cleaned: List[Tuple[int, int]] = []
+    for start, end in intervals:
+        start = max(0, min(length - 1, int(start)))
+        end = max(start + 1, min(length, int(end)))
+        cleaned.append((start, end))
+
+    return _AxisPlan(
+        intervals=cleaned,
+        cuts=cuts,
+        score=float(np.mean(scores)) if scores else 0.0,
+    )
+
+
+def _parse_layout(layout: str, custom_rows: int, custom_cols: int) -> Tuple[int, int]:
+    if layout == "custom":
+        return max(1, int(custom_rows)), max(1, int(custom_cols))
+    rows_text, cols_text = layout.split("x", 1)
+    return int(rows_text), int(cols_text)
+
+
+def _fallback_layout(width: int, height: int) -> Tuple[int, int]:
+    aspect = width / max(1, height)
+    if 0.82 <= aspect <= 1.22:
+        return 3, 3
+    if aspect > 1.22:
+        return 2, 3
+    return 3, 2
+
+
+def _choose_auto_layout(
+    arr: np.ndarray,
+    search_px: int,
+    crop_separators: bool,
+    trim_outer: bool,
+    min_separator_px: int,
+) -> Tuple[int, int, _AxisPlan, _AxisPlan, float, bool]:
+    height, width = arr.shape[:2]
+    best = None
+
+    for rows, cols in _AUTO_LAYOUTS:
+        x_plan = _axis_plan(arr, cols, "x", search_px, crop_separators, trim_outer, min_separator_px)
+        y_plan = _axis_plan(arr, rows, "y", search_px, crop_separators, trim_outer, min_separator_px)
+        score = (x_plan.score + y_plan.score) / 2.0
+
+        # Prefer common 3x3 / 2x3 / 3x2 layouts when the image gives weak signals.
+        if (rows, cols) in ((3, 3), (2, 3), (3, 2)):
+            score += 0.025
+
+        if best is None or score > best[0]:
+            best = (score, rows, cols, x_plan, y_plan)
+
+    assert best is not None
+    score, rows, cols, x_plan, y_plan = best
+    confident = score >= 0.22
+
+    if confident:
+        return rows, cols, x_plan, y_plan, score, True
+
+    rows, cols = _fallback_layout(width, height)
+    x_plan = _axis_plan(arr, cols, "x", search_px, crop_separators, trim_outer, min_separator_px)
+    y_plan = _axis_plan(arr, rows, "y", search_px, crop_separators, trim_outer, min_separator_px)
+    return rows, cols, x_plan, y_plan, score, False
+
+
+def _normalize_sizes(crops: List[Image.Image]) -> List[Image.Image]:
+    min_w = min(crop.width for crop in crops)
+    min_h = min(crop.height for crop in crops)
+    normalized = []
+    for crop in crops:
+        left = max(0, (crop.width - min_w) // 2)
+        top = max(0, (crop.height - min_h) // 2)
+        normalized.append(crop.crop((left, top, left + min_w, top + min_h)))
+    return normalized
+
+
+def _split_one(
+    image: Image.Image,
+    layout: str,
+    custom_rows: int,
+    custom_cols: int,
+    search_px: int,
+    crop_separators: bool,
+    trim_outer: bool,
+    min_separator_px: int,
+) -> Tuple[List[Image.Image], str]:
+    arr = _to_float_array(image)
+
+    if layout == "auto":
+        rows, cols, x_plan, y_plan, confidence, confident = _choose_auto_layout(
+            arr,
+            search_px=search_px,
+            crop_separators=crop_separators,
+            trim_outer=trim_outer,
+            min_separator_px=min_separator_px,
+        )
+        mode_note = "auto" if confident else "auto-low-confidence-fallback"
+    else:
+        rows, cols = _parse_layout(layout, custom_rows, custom_cols)
+        x_plan = _axis_plan(arr, cols, "x", search_px, crop_separators, trim_outer, min_separator_px)
+        y_plan = _axis_plan(arr, rows, "y", search_px, crop_separators, trim_outer, min_separator_px)
+        confidence = (x_plan.score + y_plan.score) / 2.0
+        mode_note = "manual"
+
+    crops: List[Image.Image] = []
+    for y0, y1 in y_plan.intervals:
+        for x0, x1 in x_plan.intervals:
+            crops.append(image.crop((x0, y0, x1, y1)))
+
+    crops = _normalize_sizes(crops)
+    info = (
+        f"{mode_note}: {rows}x{cols}, cells={len(crops)}, "
+        f"confidence={confidence:.3f}, "
+        f"x={x_plan.intervals}, y={y_plan.intervals}"
+    )
+    return crops, info
+
+
+class O1keyGridSplitter:
+    """Split AI-generated grid/contact-sheet images into individual cells."""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "图像": ("IMAGE",),
+                "布局": (_LAYOUTS, {"default": "auto"}),
+                "自定义行数": ("INT", {"default": 3, "min": 1, "max": 12, "step": 1}),
+                "自定义列数": ("INT", {"default": 3, "min": 1, "max": 12, "step": 1}),
+                "搜索范围px": ("INT", {"default": 32, "min": 0, "max": 256, "step": 1}),
+                "裁掉分隔线": ("BOOLEAN", {"default": True}),
+                "裁掉外边距": ("BOOLEAN", {"default": True}),
+                "最小分隔线px": ("INT", {"default": 2, "min": 0, "max": 64, "step": 1}),
+                "最大输出张数": ("INT", {"default": 16, "min": 1, "max": 144, "step": 1}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE", "STRING")
+    RETURN_NAMES = ("切割图像", "检测信息")
+    FUNCTION = "split_grid"
+    CATEGORY = "o1key/image"
+    DESCRIPTION = (
+        "智能切割 AI 生成的九宫格、六宫格等合并图。"
+        "自动模式会检测常见布局；没有明显分隔线时建议手动选择布局。"
+    )
+
+    def split_grid(
+        self,
+        图像: torch.Tensor,
+        布局: str = "auto",
+        自定义行数: int = 3,
+        自定义列数: int = 3,
+        搜索范围px: int = 32,
+        裁掉分隔线: bool = True,
+        裁掉外边距: bool = True,
+        最小分隔线px: int = 2,
+        最大输出张数: int = 16,
+    ):
+        source_images = tensor_to_pil(图像)
+        all_crops: List[Image.Image] = []
+        info_lines: List[str] = []
+
+        for batch_index, image in enumerate(source_images, start=1):
+            crops, info = _split_one(
+                image=image,
+                layout=布局,
+                custom_rows=自定义行数,
+                custom_cols=自定义列数,
+                search_px=搜索范围px,
+                crop_separators=裁掉分隔线,
+                trim_outer=裁掉外边距,
+                min_separator_px=最小分隔线px,
+            )
+            if len(crops) > 最大输出张数:
+                raise ValueError(
+                    f"合并图切割：检测到 {len(crops)} 张，超过最大输出张数 {最大输出张数}。"
+                    "请调大最大输出张数，或检查布局设置。"
+                )
+            all_crops.extend(crops)
+            info_lines.append(f"batch {batch_index}: {info}")
+
+        if not all_crops:
+            raise ValueError("合并图切割：没有生成任何切片。")
+
+        all_crops = _normalize_sizes(all_crops)
+        print("[o1key 合并图切割] " + " | ".join(info_lines))
+        return (pil_to_tensor(all_crops), "\n".join(info_lines))
@@ -0,0 +1,165 @@
+"""
+o1key Grok Image 节点
+支持 Grok Image / Grok Image Pro 模型的文生图和图生图
+"""
+
+import time
+from ..clients.grok_image_client import GrokImageClient
+from ..utils.image_utils import parse_batch_prompts
+from ..utils.config import NETWORK_ROUTE_OPTIONS
+
+try:
+    from comfy.model_management import processing_interrupted, InterruptProcessingException
+    _INTERRUPT_AVAILABLE = True
+except ImportError:
+    _INTERRUPT_AVAILABLE = False
+    processing_interrupted = lambda: False
+    InterruptProcessingException = RuntimeError
+
+_ASPECT_RATIOS = [
+    "auto", "1:1", "16:9", "9:16", "4:3", "3:4",
+    "3:2", "2:3", "2:1", "1:2", "19.5:9", "9:19.5", "20:9", "9:20",
+]
+
+
+class O1keyGrokImage:
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        optional_inputs = {}
+        for i in range(1, 4):
+            optional_inputs[f"参考图{i}"] = ("IMAGE", {
+                "tooltip": f"Optional reference image {i}",
+            })
+
+        return {
+            "required": {
+                "prompt": ("STRING", {
+                    "default": "",
+                    "multiline": True,
+                    "tooltip": "文本提示词，用 --- 独占一行分隔批量提示词",
+                }),
+            },
+            "optional": {
+                "模型": (["Grok Image", "Grok Image Pro"], {
+                    "default": "Grok Image Pro",
+                }),
+                "宽高比": (_ASPECT_RATIOS, {
+                    "default": "auto",
+                }),
+                "分辨率": (["1k", "2k"], {
+                    "default": "1k",
+                }),
+                "生图数量": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 4,
+                    "step": 1,
+                    "display": "number",
+                }),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {
+                    "default": NETWORK_ROUTE_OPTIONS[0],
+                }),
+                "seed": ("INT", {
+                    "default": 0,
+                    "min": 0,
+                    "max": 2**31 - 1,
+                    "step": 1,
+                    "display": "number",
+                    "control_after_generate": True,
+                }),
+                **optional_inputs,
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("IMAGE",)
+    FUNCTION = "generate"
+    CATEGORY = "o1key/image"
+    OUTPUT_NODE = False
+
+    def generate(
+        self,
+        prompt: str,
+        模型: str = "Grok Image Pro",
+        宽高比: str = "auto",
+        分辨率: str = "1k",
+        生图数量: int = 1,
+        网络线路: str = "全球加速",
+        seed: int = 0,
+        **kwargs,
+    ):
+        start_time = time.time()
+
+        reference_tensors = []
+        for i in range(1, 4):
+            key = f"参考图{i}"
+            if key in kwargs and kwargs[key] is not None:
+                reference_tensors.append(kwargs[key])
+        image_list = reference_tensors if reference_tensors else None
+
+        try:
+            client = GrokImageClient(route=网络线路)
+        except ValueError as e:
+            if str(e) == "未授权！":
+                print("[o1key Grok Image] 请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            raise
+
+        try:
+            batch_prompts = parse_batch_prompts(prompt)
+            all_pil_images = []
+
+            if batch_prompts:
+                total = len(batch_prompts)
+                print(f"[o1key Grok Image] 批量模式 | {total} 条提示词 | 每条生成 {生图数量} 张")
+                for idx, p in enumerate(batch_prompts, 1):
+                    if _INTERRUPT_AVAILABLE and processing_interrupted():
+                        print("[o1key Grok Image] 用户取消")
+                        raise InterruptProcessingException()
+                    try:
+                        pil_images = client.run_sync(
+                            prompt=p, model=模型, aspect_ratio=宽高比,
+                            resolution=分辨率, n=生图数量, image_list=image_list,
+                        )
+                        all_pil_images.extend(pil_images)
+                        snippet = p[:30] + ("..." if len(p) >= 30 else "")
+                        print(f"[o1key Grok Image] [{idx}/{total}] done: {snippet}")
+                    except InterruptProcessingException:
+                        raise
+                    except Exception as e:
+                        error_msg = str(e).split('\n')[0]
+                        snippet = p[:30] + ("..." if len(p) >= 30 else "")
+                        print(f"[o1key Grok Image] [{idx}/{total}] fail: {snippet} → {error_msg}")
+            else:
+                if not prompt or not prompt.strip():
+                    raise ValueError("提示词不能为空")
+                pil_images = client.run_sync(
+                    prompt=prompt, model=模型, aspect_ratio=宽高比,
+                    resolution=分辨率, n=生图数量, image_list=image_list,
+                )
+                all_pil_images.extend(pil_images)
+
+            if not all_pil_images:
+                raise RuntimeError("所有提示词均生成失败，无可用图像输出")
+
+            output_tensor = GrokImageClient._pil_list_to_tensor(all_pil_images)
+
+            elapsed = time.time() - start_time
+            print(
+                f"[o1key Grok Image] 完成！耗时 {elapsed:.1f}s，"
+                f"输出 {output_tensor.shape[0]} 张 "
+                f"{output_tensor.shape[2]}x{output_tensor.shape[1]}"
+            )
+            return (output_tensor,)
+
+        finally:
+            self._print_balance(client)
+
+    def _print_balance(self, client):
+        try:
+            balance_data = client.query_balance_sync()
+            balance_info = client.format_balance_info(balance_data)
+            print(f"[o1key Grok Image] {balance_info}")
+        except Exception:
+            pass
@@ -0,0 +1,283 @@
+"""
+Grok Video node.
+
+Submits a /v1/videos task, polls until completion, downloads the mp4,
+and returns ComfyUI's native VIDEO object.
+"""
+
+import json
+import os
+from typing import List, Optional
+
+from ..clients.grok_video_client import GrokVideoClient
+from ..utils.config import NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.image_utils import encode_images_for_request_body_limit, tensor_to_pil
+
+try:
+    import folder_paths
+    FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    FOLDER_PATHS_AVAILABLE = False
+
+try:
+    from comfy.utils import ProgressBar
+    PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    ProgressBar = None
+    PROGRESS_BAR_AVAILABLE = False
+
+try:
+    from comfy_api.input_impl import VideoFromFile
+except Exception:
+    try:
+        from comfy_api.latest import InputImpl
+        VideoFromFile = InputImpl.VideoFromFile
+    except Exception:
+        VideoFromFile = None
+
+
+MODEL_OPTIONS = ["grok-imagine-video-1.5-preview", "grok-imagine-1.0-video"]
+ASPECT_RATIO_OPTIONS = ["1:1", "16:9", "9:16", "4:3", "3:4", "3:2", "2:3"]
+QUALITY_OPTIONS = ["720p"]
+QUALITY_VALUE_MAP = {
+    "720p": "high",
+}
+MODEL_SECONDS_OPTIONS = {
+    "grok-imagine-1.0-video": [6, 10, 12, 16, 20],
+}
+
+MAX_REFERENCE_IMAGES = 3
+MAX_REQUEST_BODY_BYTES = 20 * 1024 * 1024
+
+
+def _get_output_dir() -> str:
+    if FOLDER_PATHS_AVAILABLE:
+        base = folder_paths.get_output_directory()
+    else:
+        plugin_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+        comfy_root = os.path.dirname(os.path.dirname(plugin_dir))
+        base = os.path.join(comfy_root, "output")
+
+    output_dir = os.path.join(base, "grok_video")
+    os.makedirs(output_dir, exist_ok=True)
+    return output_dir
+
+
+def _format_mb(size_bytes: int) -> str:
+    return f"{size_bytes / 1024 / 1024:.2f}MB"
+
+
+def _image_tensor_to_first_pil(image_tensor):
+    if image_tensor is None:
+        return None
+
+    pil_images = tensor_to_pil(image_tensor)
+    if not pil_images:
+        return None
+
+    image = pil_images[0]
+    if image.mode not in ("RGB", "L"):
+        image = image.convert("RGB")
+    return image
+
+
+def _collect_reference_images(**kwargs) -> List[object]:
+    images = []
+    for i in range(1, MAX_REFERENCE_IMAGES + 1):
+        image = _image_tensor_to_first_pil(kwargs.get(f"参考图{i}"))
+        if image is not None:
+            images.append(image)
+    return images
+
+
+def _to_data_urls(encoded_images) -> List[str]:
+    return [f"data:{mime};base64,{b64}" for mime, b64 in encoded_images]
+
+
+def _encode_image_data_urls(
+    images: List[object],
+    prompt: str,
+    model: str,
+    aspect_ratio: str,
+    seconds: int,
+    quality: str,
+) -> Optional[List[str]]:
+    if not images:
+        return None
+
+    def build_body(encoded_images):
+        return GrokVideoClient.build_video_body(
+            prompt=prompt,
+            model=model,
+            aspect_ratio=aspect_ratio,
+            seconds=seconds,
+            quality=quality,
+            images=_to_data_urls(encoded_images),
+        )
+
+    encoded = encode_images_for_request_body_limit(
+        images,
+        build_body=build_body,
+        max_body_bytes=MAX_REQUEST_BODY_BYTES,
+    )
+    data_urls = _to_data_urls(encoded)
+
+    return data_urls
+
+
+def _validate_request_body_size(body: dict) -> None:
+    body_size = len(json.dumps(body, ensure_ascii=False).encode("utf-8"))
+    if body_size > MAX_REQUEST_BODY_BYTES:
+        raise ValueError(
+            f"Grok Video 请求体大小 {_format_mb(body_size)} 超过 "
+            f"{_format_mb(MAX_REQUEST_BODY_BYTES)} 限制，请减少参考图片或降低图片尺寸。"
+        )
+
+
+class O1keyGrokVideo:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词": (
+                    "STRING",
+                    {
+                        "default": "",
+                        "multiline": True,
+                    },
+                ),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": NETWORK_ROUTE_OPTIONS[0]}),
+                "模型": (MODEL_OPTIONS, {"default": MODEL_OPTIONS[0]}),
+                "宽高比": (ASPECT_RATIO_OPTIONS, {"default": "16:9"}),
+                "秒数（按模型限制）": (
+                    "INT",
+                    {
+                        "default": 5,
+                        "min": 5,
+                        "max": 20,
+                        "step": 1,
+                        "display": "number",
+                    },
+                ),
+                "画质": (QUALITY_OPTIONS, {"default": "720p"}),
+            },
+            "optional": {
+                "参考图1": ("IMAGE",),
+                "参考图2": ("IMAGE",),
+                "参考图3": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Video"
+
+    DESCRIPTION = (
+        "Grok Video /v1/videos task node. Supports prompt plus up to "
+        "three image references, multiple aspect ratios, model-specific seconds, 720p output."
+    )
+
+    def generate(
+        self,
+        **kwargs,
+    ):
+        if VideoFromFile is None:
+            raise RuntimeError("当前 ComfyUI 版本不支持原生 VIDEO 输入实现 VideoFromFile。")
+
+        提示词 = kwargs.get("提示词", "")
+        网络线路 = kwargs.get("网络线路", NETWORK_ROUTE_OPTIONS[0])
+        模型 = kwargs.get("模型", MODEL_OPTIONS[0])
+        宽高比 = kwargs.get("宽高比", "16:9")
+        秒数 = kwargs.get("秒数（按模型限制）", kwargs.get("秒数（≤15s）", kwargs.get("秒数", 5)))
+        画质 = kwargs.get("画质", "720p")
+
+        prompt = (提示词 or "").strip()
+        if not prompt:
+            raise ValueError("提示词不能为空。")
+        if 模型 not in MODEL_OPTIONS:
+            raise ValueError(f"模型仅支持: {', '.join(MODEL_OPTIONS)}")
+        if 宽高比 not in ASPECT_RATIO_OPTIONS:
+            raise ValueError(f"宽高比仅支持: {', '.join(ASPECT_RATIO_OPTIONS)}。")
+        seconds = int(秒数)
+        allowed_seconds = MODEL_SECONDS_OPTIONS.get(模型)
+        if allowed_seconds is not None:
+            if seconds not in allowed_seconds:
+                raise ValueError(
+                    f"模型 {模型} 仅支持秒数: "
+                    f"{', '.join(str(s) for s in allowed_seconds)}。"
+                    "请修改为正确的秒数后再发起请求。"
+                )
+        elif seconds < 5 or seconds > 15:
+            raise ValueError("秒数仅支持 5 到 15。")
+        if 画质 not in QUALITY_OPTIONS:
+            raise ValueError("画质仅支持 720p。")
+
+        quality = QUALITY_VALUE_MAP[画质]
+        reference_images = _collect_reference_images(**kwargs)
+        image_data_urls = _encode_image_data_urls(
+            reference_images,
+            prompt=prompt,
+            model=模型,
+            aspect_ratio=宽高比,
+            seconds=seconds,
+            quality=quality,
+        )
+
+        request_body = GrokVideoClient.build_video_body(
+            prompt=prompt,
+            model=模型,
+            aspect_ratio=宽高比,
+            seconds=seconds,
+            quality=quality,
+            images=image_data_urls,
+        )
+        _validate_request_body_size(request_body)
+
+        pbar = ProgressBar(100) if PROGRESS_BAR_AVAILABLE else None
+        last_progress = [0]
+
+        def progress_callback(progress: int, status: str, elapsed: float):
+            progress_value = max(0, min(100, int(progress or 0)))
+            if pbar is not None and progress_value > last_progress[0]:
+                pbar.update(progress_value - last_progress[0])
+                last_progress[0] = progress_value
+
+        client = GrokVideoClient(base_url=get_base_url_by_route(网络线路))
+
+        try:
+            result = client.generate_video_sync(
+                prompt=prompt,
+                model=模型,
+                aspect_ratio=宽高比,
+                seconds=seconds,
+                quality=quality,
+                output_dir=_get_output_dir(),
+                images=image_data_urls,
+                poll_interval=5,
+                timeout=1200,
+                progress_callback=progress_callback,
+            )
+
+            if pbar is not None and last_progress[0] < 100:
+                pbar.update(100 - last_progress[0])
+
+            video_path = result["video_path"]
+            print(f"Grok Video：下载完成：{video_path}")
+            return (VideoFromFile(video_path),)
+        finally:
+            try:
+                balance_data = client.query_balance_sync()
+                balance_info = client.format_balance_info(balance_data)
+                print(f"Grok Video：{balance_info}")
+            except Exception:
+                pass
+
+
+NODE_CLASS_MAPPINGS = {
+    "O1keyGrokVideo": O1keyGrokVideo,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "O1keyGrokVideo": "Grok Video",
+}
@@ -0,0 +1,241 @@
+"""
+高级图像拼接节点
+支持最多 10 张图像按指定方向（上、下、左、右）依次拼接，
+支持调整图像大小匹配和添加间隔。
+"""
+
+from typing import Optional, Tuple, List
+
+import torch
+from PIL import Image
+
+from ..utils.image_utils import tensor_to_pil, pil_to_tensor
+from ..utils.file_utils import load_images_from_folder
+
+
+# 间隔颜色映射
+SPACING_COLOR_MAP = {
+    "white": (255, 255, 255),
+    "black": (0, 0, 0),
+    "red": (255, 0, 0),
+    "green": (0, 255, 0),
+    "blue": (0, 0, 255),
+}
+
+
+def _resize_to_match(img: Image.Image, ref: Image.Image, direction: str) -> Image.Image:
+    """
+    按拼接方向将 img 缩放，使其与 ref 在垂直于拼接轴的尺寸上一致。
+
+    - 水平拼接 (right/left)：统一高度
+    - 垂直拼接 (down/up)：统一宽度
+    """
+    ref_w, ref_h = ref.size
+    img_w, img_h = img.size
+
+    if direction in ("right", "left"):
+        if img_h != ref_h:
+            scale = ref_h / img_h
+            new_w = max(1, int(img_w * scale))
+            img = img.resize((new_w, ref_h), Image.LANCZOS)
+    else:
+        if img_w != ref_w:
+            scale = ref_w / img_w
+            new_h = max(1, int(img_h * scale))
+            img = img.resize((ref_w, new_h), Image.LANCZOS)
+
+    return img
+
+
+def _make_spacer(ref: Image.Image, spacing_width: int,
+                 direction: str, color: Tuple[int, int, int]) -> Image.Image:
+    """创建间隔色块"""
+    if direction in ("right", "left"):
+        return Image.new("RGB", (spacing_width, ref.size[1]), color)
+    else:
+        return Image.new("RGB", (ref.size[0], spacing_width), color)
+
+
+def _stitch_two(img_a: Image.Image, img_b: Image.Image,
+                direction: str, match_size: bool,
+                spacing_width: int, spacing_color: Tuple[int, int, int]) -> Image.Image:
+    """
+    将两张 PIL 图像按指定方向拼接。
+    img_a 为基准图像，img_b 拼接在 img_a 的指定方向侧。
+    direction="right" → img_b 在 img_a 右侧
+    direction="left"  → img_b 在 img_a 左侧
+    direction="down"  → img_b 在 img_a 下方
+    direction="up"    → img_b 在 img_a 上方
+    """
+    if img_a.mode != "RGB":
+        img_a = img_a.convert("RGB")
+    if img_b.mode != "RGB":
+        img_b = img_b.convert("RGB")
+
+    if match_size:
+        img_b = _resize_to_match(img_b, img_a, direction)
+
+    if direction == "right":
+        pieces = [img_a, img_b]
+    elif direction == "left":
+        pieces = [img_b, img_a]
+    elif direction == "down":
+        pieces = [img_a, img_b]
+    else:  # up
+        pieces = [img_b, img_a]
+
+    if spacing_width > 0:
+        interleaved: List[Image.Image] = []
+        for idx, piece in enumerate(pieces):
+            interleaved.append(piece)
+            if idx < len(pieces) - 1:
+                interleaved.append(_make_spacer(piece, spacing_width, direction, spacing_color))
+        pieces = interleaved
+
+    if direction in ("right", "left"):
+        total_w = sum(p.size[0] for p in pieces)
+        max_h = max(p.size[1] for p in pieces)
+        canvas = Image.new("RGB", (total_w, max_h), spacing_color)
+        x = 0
+        for piece in pieces:
+            canvas.paste(piece, (x, 0))
+            x += piece.size[0]
+    else:
+        max_w = max(p.size[0] for p in pieces)
+        total_h = sum(p.size[1] for p in pieces)
+        canvas = Image.new("RGB", (max_w, total_h), spacing_color)
+        y = 0
+        for piece in pieces:
+            canvas.paste(piece, (0, y))
+            y += piece.size[1]
+
+    return canvas
+
+
+def _natural_sort_key(filename: str):
+    """按数字优先的文件名排序，使 1, 2, 3, 10 而非 1, 10, 2, 3"""
+    try:
+        return (0, int(filename))
+    except ValueError:
+        return (1, filename.lower())
+
+
+class ImageStitchPro:
+    """
+    高级图像拼接节点
+
+    在 ComfyUI 原生拼接节点基础上扩展，支持同时输入最多 10 张图像，
+    按指定方向依次拼接，并可在图像间添加任意颜色的间隔。
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "方向": (["right", "down", "left", "up"], {"default": "down"}),
+                "匹配图像尺寸": ("BOOLEAN", {"default": True}),
+                "间距宽度": ("INT", {"default": 0, "min": 0, "max": 1024, "step": 2}),
+                "间距颜色": (["white", "black", "red", "green", "blue"], {"default": "white"}),
+            },
+            "optional": {
+                "图1": ("IMAGE",),
+                "图2":  ("IMAGE",),
+                "图3":  ("IMAGE",),
+                "图4":  ("IMAGE",),
+                "图5":  ("IMAGE",),
+                "图6":  ("IMAGE",),
+                "图7":  ("IMAGE",),
+                "图8":  ("IMAGE",),
+                "图9":  ("IMAGE",),
+                "图10": ("IMAGE",),
+                "图11": ("IMAGE",),
+                "图12": ("IMAGE",),
+                "图片路径（可选）": ("STRING", {"default": "", "multiline": False}),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("拼接图像",)
+    FUNCTION = "stitch"
+    CATEGORY = "image"
+
+    DESCRIPTION = (
+        "高级图像拼接节点，支持最多 12 张图像按指定方向（右/下/左/上）依次拼接。\n"
+        "可选择是否将后续图像缩放以匹配第一张图像的尺寸，并可在图像间添加彩色间隔。\n"
+        "可选填「图片路径」：仅处理该文件夹内图片，按文件名顺序依次拼接；与输入端图片不可同时使用。"
+    )
+
+    def stitch(
+        self,
+        方向: str = "down",
+        匹配图像尺寸: bool = True,
+        间距宽度: int = 0,
+        间距颜色: str = "white",
+        图1:  Optional[torch.Tensor] = None,
+        图2:  Optional[torch.Tensor] = None,
+        图3:  Optional[torch.Tensor] = None,
+        图4:  Optional[torch.Tensor] = None,
+        图5:  Optional[torch.Tensor] = None,
+        图6:  Optional[torch.Tensor] = None,
+        图7:  Optional[torch.Tensor] = None,
+        图8:  Optional[torch.Tensor] = None,
+        图9:  Optional[torch.Tensor] = None,
+        图10: Optional[torch.Tensor] = None,
+        图11: Optional[torch.Tensor] = None,
+        图12: Optional[torch.Tensor] = None,
+        **kwargs: object,
+    ) -> Tuple[torch.Tensor]:
+
+        color = SPACING_COLOR_MAP.get(间距颜色, (255, 255, 255))
+        raw_tensors = [图1, 图2, 图3, 图4, 图5, 图6, 图7, 图8, 图9, 图10, 图11, 图12]
+        tensors = [t for t in raw_tensors if t is not None]
+        has_input_images = len(tensors) > 0
+        image_folder = (kwargs.get("图片路径（可选）") or "").strip()
+
+        if image_folder and has_input_images:
+            raise ValueError("不可同时使用「图片路径（可选）」与输入端图片，请二选一。")
+
+        if image_folder:
+            infos = load_images_from_folder(image_folder)
+            if not infos:
+                raise ValueError(f"文件夹中未找到可用的图片，或路径无效: {image_folder}")
+            infos.sort(key=lambda x: _natural_sort_key(x.filename))
+            pil_list = [info.image for info in infos]
+            if len(pil_list) == 1:
+                return (pil_to_tensor(pil_list),)
+            base = pil_list[0]
+            for next_img in pil_list[1:]:
+                base = _stitch_two(
+                    base, next_img,
+                    direction=方向,
+                    match_size=匹配图像尺寸,
+                    spacing_width=间距宽度,
+                    spacing_color=color,
+                )
+            return (pil_to_tensor([base]),)
+        else:
+            if not has_input_images:
+                raise ValueError("请至少接入一张图片，或填写「图片路径（可选）」中的文件夹路径。")
+
+        if len(tensors) == 1:
+            return (tensors[0],)
+
+        pil_batches: List[List[Image.Image]] = [tensor_to_pil(t) for t in tensors]
+
+        batch_size = min(len(b) for b in pil_batches)
+        result_images: List[Image.Image] = []
+
+        for i in range(batch_size):
+            frames = [batch[i] for batch in pil_batches]
+            base = frames[0]
+            for next_img in frames[1:]:
+                base = _stitch_two(
+                    base, next_img,
+                    direction=方向,
+                    match_size=匹配图像尺寸,
+                    spacing_width=间距宽度,
+                    spacing_color=color,
+                )
+            result_images.append(base)
+
+        return (pil_to_tensor(result_images),)
@@ -0,0 +1,738 @@
+"""
+Kling 3.0 Video Nodes
+"""
+
+import os
+import tempfile
+
+from ..clients.kling_client import KlingClient
+from ..clients.gemini_client import GeminiAPIClient
+from ..utils.image_utils import tensor_to_pil, encode_image_to_base64
+from ..utils.config import NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+
+from comfy_api.latest import InputImpl
+
+
+def _tensor_to_base64(tensor) -> str:
+    """ComfyUI IMAGE tensor → base64 PNG 字符串"""
+    pil_images = tensor_to_pil(tensor)
+    return encode_image_to_base64(pil_images[0], format="PNG")
+
+
+def _validate_prompt(prompt: str, *, required: bool = True) -> None:
+    """校验单条提示词。
+
+    Args:
+        prompt:   提示词字符串。
+        required: 为 True 时不允许为空（多镜头关闭或 shot_type 为 intelligence 时适用）。
+    """
+    if required and not prompt.strip():
+        raise ValueError("提示词不能为空（非多镜头模式下必填）。")
+    if len(prompt) > 2500:
+        raise ValueError(
+            f"提示词长度 ({len(prompt)}) 超过上限 2500 个字符，请缩短后重试。"
+        )
+
+
+def _validate_multi_prompt(multi_prompt_list: list, total_duration: int) -> None:
+    """校验多镜头分镜列表。
+
+    规则：
+    - 分镜数量：1 ~ 6；
+    - 每个分镜提示词不超过 512 个字符；
+    - 每个分镜时长 ≥ 1 且 ≤ total_duration；
+    - 所有分镜时长之和必须等于 total_duration。
+    """
+    count = len(multi_prompt_list)
+    if count < 1 or count > 6:
+        raise ValueError(
+            f"多镜头分镜数量须在 1~6 之间，当前为 {count}。"
+        )
+
+    duration_sum = 0
+    for entry in multi_prompt_list:
+        idx      = entry["index"]
+        p        = entry.get("prompt", "")
+        dur      = entry.get("duration", 0)
+
+        if len(p) > 512:
+            raise ValueError(
+                f"镜头 {idx} 提示词长度 ({len(p)}) 超过上限 512 个字符。"
+            )
+        if dur < 1:
+            raise ValueError(
+                f"镜头 {idx} 时长 ({dur}s) 不能小于 1 秒。"
+            )
+        if dur > total_duration:
+            raise ValueError(
+                f"镜头 {idx} 时长 ({dur}s) 超过任务总时长 ({total_duration}s)。"
+            )
+        duration_sum += dur
+
+    if duration_sum != total_duration:
+        raise ValueError(
+            f"所有分镜时长之和 ({duration_sum}s) 必须等于任务总时长 ({total_duration}s)。"
+        )
+
+
+def _validate_image(tensor, label: str = "图片") -> None:
+    """校验图片张量。
+
+    规则：
+    - 文件大小（PNG）不超过 10MB；
+    - 宽、高均不小于 300px；
+    - 宽高比介于 1:2.5 ~ 2.5:1 之间（即 ratio ∈ [0.4, 2.5]）。
+    """
+    import io
+
+    pil_images = tensor_to_pil(tensor)
+    img = pil_images[0]
+    w, h = img.size
+
+    # ── 最小尺寸 ──────────────────────────────────────────────────────
+    if w < 300 or h < 300:
+        raise ValueError(
+            f"{label} 宽高不得小于 300px，当前为 {w}×{h}px。"
+        )
+
+    # ── 宽高比 ────────────────────────────────────────────────────────
+    ratio = w / h
+    if ratio < 1 / 2.5 or ratio > 2.5:
+        raise ValueError(
+            f"{label} 宽高比须在 1:2.5 ~ 2.5:1 之间，"
+            f"当前为 {w}:{h}（比值 {ratio:.2f}）。"
+        )
+
+    # ── 文件大小 ──────────────────────────────────────────────────────
+    buf = io.BytesIO()
+    img.save(buf, format="PNG")
+    size_mb = buf.tell() / (1024 * 1024)
+    if size_mb > 10:
+        raise ValueError(
+            f"{label} PNG 大小 ({size_mb:.1f}MB) 超过上限 10MB。"
+        )
+
+
+class KlingVideo:
+    """Kling 视频生成节点（支持多镜头）"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词": ("STRING", {"multiline": True, "default": ""}),
+                "反向提示词": ("STRING", {"multiline": True, "default": ""}),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型版本": (["v3", "v2-6"], {"default": "v3"}),
+                "时长": ([5, 10, 15],),
+                "分辨率": (["1080p", "720p"],),
+                "宽高比": (["智能", "16:9", "9:16", "1:1"], {"default": "智能"}),
+                "生成音频": (["打开", "关闭"], {"default": "打开"}),
+                "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffff}),
+            },
+            "optional": {
+                "起始帧": ("IMAGE",),
+                "镜头1_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头1_时长": ("STRING", {"default": "5"}),
+                "镜头2_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头2_时长": ("STRING", {"default": "5"}),
+                "镜头3_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头3_时长": ("STRING", {"default": "5"}),
+                "镜头4_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头4_时长": ("STRING", {"default": "5"}),
+                "镜头5_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头5_时长": ("STRING", {"default": "5"}),
+                "镜头6_提示词": ("STRING", {"multiline": True, "default": ""}),
+                "镜头6_时长": ("STRING", {"default": "5"}),
+            }
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Kling"
+
+    async def generate(self, **kwargs):
+        """生成视频（支持多镜头）"""
+        prompt          = kwargs["提示词"]
+        negative_prompt = kwargs["反向提示词"]
+        model_ver       = kwargs.get("模型版本", "v3")
+        duration        = kwargs["时长"]
+        resolution      = kwargs["分辨率"]
+        aspect_ratio    = kwargs["宽高比"]
+        generate_audio  = kwargs["生成音频"]
+        start_frame     = kwargs.get("起始帧", None)
+        seed            = kwargs.get("seed", 0)  # noqa: F841 — 触发 ComfyUI 缓存刷新
+
+        mode  = "pro" if resolution == "1080p" else "std"
+        voice = "voice" if generate_audio == "打开" else "novoice"
+
+        # ── v2-6 模型约束校验 ──────────────────────────────────────────
+        if model_ver == "v2-6":
+            if duration == 15:
+                raise ValueError(
+                    "v2-6 模型不支持 15s 时长，请选择 5s 或 10s。"
+                )
+            if mode == "std" and voice == "voice":
+                raise ValueError(
+                    "v2-6 模型的标准画质（720p）不支持生成音频，请关闭生成音频或切换至 1080p。"
+                )
+
+        # ── 多镜头检测 ────────────────────────────────────────────────
+        multi_prompt_list = []
+        for i in range(1, 7):
+            sb_prompt = kwargs.get(f"镜头{i}_提示词", "").strip()
+            if sb_prompt:
+                raw_dur = kwargs.get(f"镜头{i}_时长", "5")
+                try:
+                    sb_duration = int(str(raw_dur).strip()) if str(raw_dur).strip() else 5
+                except ValueError:
+                    sb_duration = 5
+                multi_prompt_list.append({
+                    "index": i,
+                    "prompt": sb_prompt,
+                    "duration": sb_duration,
+                })
+
+        multi_shot_enabled = len(multi_prompt_list) > 0
+
+        if multi_shot_enabled:
+            total_duration = sum(e["duration"] for e in multi_prompt_list)
+            if total_duration < 3 or total_duration > 15:
+                raise ValueError(
+                    f"多镜头总时长 ({total_duration}s) 必须在 3~15 秒之间。"
+                )
+            _validate_multi_prompt(multi_prompt_list, total_duration)
+            duration = total_duration
+        else:
+            _validate_prompt(prompt, required=True)
+
+        # ── 构建模型名 & 请求体 ───────────────────────────────────────
+        import json, base64, copy
+        model_name = f"kling-{model_ver}-{mode}-{duration}s-{voice}"
+
+        body = {
+            "model":    model_name,
+            "mode":     mode,
+            "duration": duration,
+        }
+
+        sound = "on" if generate_audio == "打开" else "off"
+
+        if multi_shot_enabled or sound == "on":
+            ms_payload = {}
+            ms_payload["prompt"] = prompt
+
+            if sound == "on":
+                ms_payload["sound"] = "on"
+
+            if multi_shot_enabled:
+                ms_payload["multi_shot"] = True
+                ms_payload["shot_type"] = "customize"
+                ms_payload["multi_prompt"] = multi_prompt_list
+
+            encoded = base64.b64encode(
+                json.dumps(ms_payload, ensure_ascii=False).encode("utf-8")
+            ).decode("utf-8")
+            body["prompt"] = f"__MS__:{encoded}"
+        else:
+            body["prompt"] = prompt
+
+        if negative_prompt.strip():
+            body["negative_prompt"] = negative_prompt
+
+        if start_frame is not None:
+            _validate_image(start_frame, "起始帧")
+            body["image"] = _tensor_to_base64(start_frame)
+            endpoint_type = "image2video"
+        else:
+            if aspect_ratio and aspect_ratio != "智能":
+                body["metadata"] = {"aspect_ratio": aspect_ratio}
+            endpoint_type = "text2video"
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        _, save_path = tempfile.mkstemp(suffix=".mp4", prefix="kling_")
+
+        client = KlingClient()
+        client.base_url = get_base_url_by_route(kwargs.get("网络线路", "全球加速"))
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def on_stage(stage: str):
+            if stage == "submitting":
+                print("[视频生成] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif stage.startswith("submitted:"):
+                print(f"[视频生成] 任务已提交 → {stage.split(':',1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif stage == "downloading":
+                print("[视频生成] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif stage == "done":
+                print("[视频生成] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def on_progress(pct: int):
+            mapped = 5 + int(pct * 0.94)
+            if pbar: pbar.update_absolute(mapped, 100)
+
+        try:
+            result_path = await client.generate_async(
+                endpoint_type=endpoint_type,
+                body=body,
+                save_path=save_path,
+                on_stage=on_stage,
+                on_progress=on_progress,
+            )
+            return (InputImpl.VideoFromFile(result_path),)
+        finally:
+            # 查询余额
+            try:
+                _balance_client = GeminiAPIClient()
+                balance_data = _balance_client.query_balance_sync()
+                balance_info = _balance_client.format_balance_info(balance_data)
+                print(f"自研视频模型: {balance_info}")
+            except Exception:
+                pass
+
+
+class KlingFirstLastFrame:
+    """Kling 首尾帧到视频节点"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "首帧": ("IMAGE",),
+                "尾帧": ("IMAGE",),
+                "提示词": ("STRING", {"multiline": True, "default": ""}),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型": (["v3", "v2-6"], {"default": "v3"}),
+                "分辨率": (["1080p", "720p"],),
+                "时长": ([5, 10, 15],),
+                "生成音频": (["打开", "关闭"], {"default": "打开"}),
+                "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffff}),
+            }
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Kling"
+
+    async def generate(self, **kwargs):
+        first_frame  = kwargs["首帧"]
+        end_frame    = kwargs["尾帧"]
+        prompt       = kwargs["提示词"]
+        duration     = kwargs["时长"]
+        generate_audio = kwargs["生成音频"]
+        model_base   = kwargs["模型"]
+        model_base   = "kling-" + model_base          # v3/v2-6 → kling-v3/kling-v2-6（后端值还原）
+        resolution   = kwargs["分辨率"]
+        seed         = kwargs.get("seed", 0)  # noqa: F841 — 触发 ComfyUI 缓存刷新
+
+        _validate_prompt(prompt, required=True)
+
+        # 时长校验
+        if duration not in (5, 10, 15):
+            raise ValueError(f"时长仅支持 5、10、15 秒，当前值为 {duration}，请重新选择。")
+
+        # 拼接模型名：kling-{ver}-{mode}-{dur}s-{voice}
+        mode  = "pro" if resolution == "1080p" else "std"
+        voice = "voice" if generate_audio == "打开" else "novoice"
+
+        # ── v2-6 模型约束校验 ──────────────────────────────────────────
+        model_ver = kwargs["模型"]   # "v3" or "v2-6"
+        if model_ver == "v2-6":
+            if duration == 15:
+                raise ValueError(
+                    "v2-6 模型不支持 15s 时长，请选择 5s 或 10s。"
+                )
+            if mode == "std" and voice == "voice":
+                raise ValueError(
+                    "v2-6 模型的标准画质（720p）不支持生成音频，请关闭生成音频或切换至 1080p。"
+                )
+
+        model_name = f"{model_base}-{mode}-{duration}s-{voice}"
+
+        # 图片校验 & 转 base64
+        _validate_image(first_frame, "首帧")
+        _validate_image(end_frame,   "尾帧")
+        image_b64      = _tensor_to_base64(first_frame)
+        image_tail_b64 = _tensor_to_base64(end_frame)
+
+        # ── 按规范编码 prompt 和 sound ──────────────────────────
+        import json, base64
+        sound = "on" if generate_audio == "打开" else "off"
+
+        body = {
+            "model":    model_name,
+            "image":    image_b64,
+            "mode":     mode,
+            "duration": duration,
+            "metadata": {
+                "image_tail": image_tail_b64,
+            },
+        }
+
+        if sound == "on":
+            ms_payload = {
+                "prompt": prompt,
+                "sound": "on",
+            }
+            encoded = base64.b64encode(
+                json.dumps(ms_payload, ensure_ascii=False).encode("utf-8")
+            ).decode("utf-8")
+            body["prompt"] = f"__MS__:{encoded}"
+        else:
+            body["prompt"] = prompt
+
+        # 保存路径（临时文件，避免与下游保存节点重复落盘）
+        _, save_path = tempfile.mkstemp(suffix=".mp4", prefix="kling_")
+
+        client = KlingClient()
+        client.base_url = get_base_url_by_route(kwargs.get("网络线路", "全球加速"))
+
+        # 进度条：0~100 步
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def on_stage(stage: str):
+            if stage == "submitting":
+                print("[视频生成] 提交中...")
+                if pbar:
+                    pbar.update_absolute(0, 100)
+            elif stage.startswith("submitted:"):
+                print(f"[视频生成] 任务已提交 → {stage.split(':',1)[1]}")
+                if pbar:
+                    pbar.update_absolute(5, 100)
+            elif stage == "downloading":
+                print("[视频生成] 下载视频...")
+                if pbar:
+                    pbar.update_absolute(99, 100)
+            elif stage == "done":
+                print("[视频生成] 完成")
+                if pbar:
+                    pbar.update_absolute(100, 100)
+
+        def on_progress(pct: int):
+            # pct 来自 API progress 字段，如 50 表示 50%
+            # 生成阶段占 5~99 区间
+            mapped = 5 + int(pct * 0.94)
+            if pbar:
+                pbar.update_absolute(mapped, 100)
+
+        try:
+            result_path = await client.generate_async(
+                endpoint_type="image2video",
+                body=body,
+                save_path=save_path,
+                on_stage=on_stage,
+                on_progress=on_progress,
+            )
+            return (InputImpl.VideoFromFile(result_path),)
+        finally:
+            # 查询余额
+            try:
+                _balance_client = GeminiAPIClient()
+                balance_data = _balance_client.query_balance_sync()
+                balance_info = _balance_client.format_balance_info(balance_data)
+                print(f"自研视频模型: {balance_info}")
+            except Exception:
+                pass
+
+
+class KlingMotionControlTest:
+    """Kling 动作控制（测试）节点 —— reference_video 接受 VIDEO 类型输入"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词":     ("STRING", {"multiline": True, "default": ""}),
+                "参考图片":   ("IMAGE",),
+                "参考视频":   ("VIDEO",),
+                "网络线路":   (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+            },
+            "optional": {
+                "模型":       (["v3", "v2-6"], {"default": "v3"}),
+                "分辨率":     (["1080p", "720p"],),
+                "时长":       ([5, 10, 15], {"default": 5}),
+                "人物朝向":   (["video", "image"],),
+                "保留原声":   (["打开", "关闭"], {"default": "打开"}),
+                "seed":       ("INT", {"default": 0, "min": 0, "max": 0xffffffff}),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Kling"
+
+    async def generate(self, **kwargs):
+        """动作控制：VIDEO 类型参考视频 + 图片人物动作迁移（走 new API 三段式）"""
+        import base64
+
+        prompt                = kwargs["提示词"]
+        reference_image       = kwargs["参考图片"]
+        reference_video       = kwargs["参考视频"]
+        keep_original_sound   = kwargs.get("保留原声", "打开")
+        character_orientation = kwargs.get("人物朝向", "video")
+        mode                  = kwargs.get("分辨率", "1080p")
+        duration              = kwargs.get("时长", 5)
+        mode_api              = "pro" if mode == "1080p" else "std"  # 映射为 API 参数值
+        model                 = kwargs.get("模型", "v3")
+        model_name            = f"kling-{model}-motion-{mode_api}-{duration}s"
+        seed                  = kwargs.get("seed", 0)  # noqa: F841 — 触发 ComfyUI 缓存刷新
+
+        # ── 校验提示词 ────────────────────────────────────────────────
+        _validate_prompt(prompt, required=True)
+
+        # ── 校验参考图片 ──────────────────────────────────────────────
+        _validate_image(reference_image, "参考图片")
+        image_b64 = _tensor_to_base64(reference_image)
+
+        # ── 从 VIDEO 对象获取本地文件路径并读取 ───────────────────────
+        video_path = None
+        if hasattr(reference_video, "source_path"):
+            video_path = reference_video.source_path
+        elif hasattr(reference_video, "path"):
+            video_path = reference_video.path
+        elif isinstance(reference_video, str):
+            video_path = reference_video.strip()
+
+        if not video_path or not os.path.isfile(video_path):
+            raise ValueError(
+                f"无法获取参考视频文件路径，请确保连接的是本地视频文件。"
+                f"（当前路径：{video_path}）"
+            )
+
+        # ── 校验视频时长约束 ──────────────────────────────────────────
+        # 人物朝向="video" → 3~30 秒；人物朝向="image" → 3~10 秒
+        try:
+            import subprocess, json as _json
+            ffprobe_cmd = [
+                "ffprobe", "-v", "quiet",
+                "-print_format", "json",
+                "-show_format",
+                video_path,
+            ]
+            result_proc = subprocess.run(ffprobe_cmd, capture_output=True, text=True, timeout=30)
+            if result_proc.returncode == 0:
+                info = _json.loads(result_proc.stdout)
+                duration_sec = float(info.get("format", {}).get("duration", 0))
+                if character_orientation == "video":
+                    if not (3 <= duration_sec <= 30):
+                        raise ValueError(
+                            f"当人物朝向为 'video' 时，"
+                            f"参考视频时长须在 3~30 秒之间，当前为 {duration_sec:.1f}s。"
+                        )
+                else:  # "image"
+                    if not (3 <= duration_sec <= 10):
+                        raise ValueError(
+                            f"当人物朝向为 'image' 时，"
+                            f"参考视频时长须在 3~10 秒之间，当前为 {duration_sec:.1f}s。"
+                        )
+        except FileNotFoundError:
+            print("[动作控制] 警告：ffprobe 未找到，跳过视频时长校验。")
+        except ValueError:
+            raise
+        except Exception as e:
+            print(f"[动作控制] 时长校验异常（已跳过）：{e}")
+
+        # ── 视频转 base64 ─────────────────────────────────────────────
+        with open(video_path, "rb") as f:
+            video_b64 = base64.b64encode(f.read()).decode("utf-8")
+
+        # ── 构建请求体（new API 格式）─────────────────────────────────
+        body = {
+            "model":                 model_name,
+            "prompt":                prompt,
+            "image_url":             image_b64,
+            "video_url":             video_b64,
+            "character_orientation": character_orientation,
+            "mode":                  mode_api,
+            "keep_original_sound":   "yes" if keep_original_sound == "打开" else "no",
+        }
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        _, save_path = tempfile.mkstemp(suffix=".mp4", prefix="kling_motion_")
+
+        client = KlingClient()
+        client.base_url = get_base_url_by_route(kwargs.get("网络线路", "全球加速"))
+
+        # ── 进度条 ────────────────────────────────────────────────────
+        try:
+            from comfy.utils import ProgressBar
+            pbar = ProgressBar(100)
+        except Exception:
+            pbar = None
+
+        def on_stage(stage: str):
+            if stage == "submitting":
+                print("[动作控制] 提交中...")
+                if pbar: pbar.update_absolute(0, 100)
+            elif stage.startswith("submitted:"):
+                print(f"[动作控制] 任务已提交 → {stage.split(':',1)[1]}")
+                if pbar: pbar.update_absolute(5, 100)
+            elif stage == "downloading":
+                print("[动作控制] 下载视频...")
+                if pbar: pbar.update_absolute(99, 100)
+            elif stage == "done":
+                print("[动作控制] 完成")
+                if pbar: pbar.update_absolute(100, 100)
+
+        def on_progress(pct: int):
+            mapped = 5 + int(pct * 0.94)
+            if pbar: pbar.update_absolute(mapped, 100)
+
+        try:
+            result_path = await client.motion_control_async(
+                body=body,
+                save_path=save_path,
+                on_stage=on_stage,
+                on_progress=on_progress,
+            )
+            return (InputImpl.VideoFromFile(result_path),)
+        finally:
+            # 查询余额
+            try:
+                _balance_client = GeminiAPIClient()
+                balance_data = _balance_client.query_balance_sync()
+                balance_info = _balance_client.format_balance_info(balance_data)
+                print(f"自研视频模型: {balance_info}")
+            except Exception:
+                pass
+
+
+class AspectRatioPreset:
+    """图片宽高比预设节点"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "图像": ("IMAGE",),
+                "宽高比": (["智能", "16:9", "9:16", "4:3", "3:4", "1:1"], {"default": "智能"}),
+            }
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("图像",)
+    FUNCTION = "resize"
+    CATEGORY = "comfyui_o1key/Utils"
+
+    def resize(self, 图像, 宽高比):
+        import torch
+        from PIL import Image
+        import numpy as np
+
+        pil_images = tensor_to_pil(图像)
+        img = pil_images[0]
+        w, h = img.size
+        img_ratio = w / h
+
+        # 确定原图所属的宽高比家族
+        ratios = {"16:9": 16/9, "9:16": 9/16, "4:3": 4/3, "3:4": 3/4, "1:1": 1.0}
+        closest_ratio = min(ratios.keys(), key=lambda k: abs(ratios[k] - img_ratio))
+
+        # 智能模式：使用最接近的比例
+        if 宽高比 == "智能":
+            宽高比 = closest_ratio
+
+        # 解析目标比例
+        target_w, target_h = map(int, 宽高比.split(":"))
+        target_ratio = target_w / target_h
+
+        # 确定分辨率级别（1K/2K）
+        max_dim = max(w, h)
+        if max_dim <= 1080:
+            base = 1080
+        elif max_dim <= 2160:
+            base = 2160
+        else:
+            base = 2160
+
+        # 计算目标尺寸
+        if target_ratio >= 1:
+            target_width = base
+            target_height = int(base / target_ratio)
+        else:
+            target_height = base
+            target_width = int(base * target_ratio)
+
+        # 判断是否同家族（横向家族：16:9, 4:3；纵向家族：9:16, 3:4；正方形：1:1）
+        horizontal_family = ["16:9", "4:3"]
+        vertical_family = ["9:16", "3:4"]
+
+        same_family = False
+        if closest_ratio in horizontal_family and 宽高比 in horizontal_family:
+            same_family = True
+        elif closest_ratio in vertical_family and 宽高比 in vertical_family:
+            same_family = True
+        elif closest_ratio == "1:1" and 宽高比 == "1:1":
+            same_family = True
+
+        # 同家族：直接缩放或裁剪（无白底）
+        if same_family:
+            if img_ratio > target_ratio:
+                # 图像更宽，以高度为准缩放后裁剪
+                scale = target_height / h
+                scaled_w = int(w * scale)
+                scaled_h = target_height
+                scaled = img.resize((scaled_w, scaled_h), Image.LANCZOS)
+                left = (scaled_w - target_width) // 2
+                result = scaled.crop((left, 0, left + target_width, target_height))
+            else:
+                # 图像更高，以宽度为准缩放后裁剪
+                scale = target_width / w
+                scaled_w = target_width
+                scaled_h = int(h * scale)
+                scaled = img.resize((scaled_w, scaled_h), Image.LANCZOS)
+                top = (scaled_h - target_height) // 2
+                result = scaled.crop((0, top, target_width, top + target_height))
+
+        # 不同家族：保持宽高比 + 白底填充
+        else:
+            if img_ratio > target_ratio:
+                scaled_w = target_width
+                scaled_h = int(target_width / img_ratio)
+            else:
+                scaled_h = target_height
+                scaled_w = int(target_height * img_ratio)
+
+            scaled = img.resize((scaled_w, scaled_h), Image.LANCZOS)
+            canvas = Image.new("RGB", (target_width, target_height), (255, 255, 255))
+            paste_x = (target_width - scaled_w) // 2
+            paste_y = (target_height - scaled_h) // 2
+            canvas.paste(scaled, (paste_x, paste_y))
+            result = canvas
+
+        # 转回 tensor
+        arr = np.array(result).astype(np.float32) / 255.0
+        tensor = torch.from_numpy(arr).unsqueeze(0)
+
+        return (tensor,)
+
+
+NODE_CLASS_MAPPINGS = {
+    "KlingVideo": KlingVideo,
+    "KlingFirstLastFrame": KlingFirstLastFrame,
+    "KlingMotionControlTest": KlingMotionControlTest,
+    "AspectRatioPreset": AspectRatioPreset,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "KlingVideo": "文/图生视频 自研模型",
+    "KlingFirstLastFrame": "首尾帧生视频 自研模型",
+    "KlingMotionControlTest": "动作控制 自研模型",
+    "AspectRatioPreset": "图片宽高比预设",
+}
@@ -0,0 +1,125 @@
+"""
+LoadFile 节点（增强版）
+支持单文件路径和文件夹路径，输出 FILE_LIST 类型供全能LLM等节点使用
+"""
+
+import base64
+import os
+from pathlib import Path
+from typing import Tuple, List
+
+from ..utils.file_types import FileData, FileList, DOCUMENT_MIME_TYPES, FILE_SIZE_LIMIT, TOTAL_FILE_SIZE_LIMIT
+
+
+class LoadFile:
+    """
+    加载文件节点
+
+    - 单文件路径：加载指定文件
+    - 文件夹路径：加载文件夹内所有支持的文件（非递归）
+    - 两者可同时使用，结果合并输出
+    - 输出 FILE_LIST 类型，可直接连接到全能LLM对话助手
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {},
+            "optional": {
+                "单文件路径": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                    "placeholder": "文件完整路径，多个文件用英文逗号分隔",
+                }),
+                "文件夹路径": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                    "placeholder": "文件夹路径，自动读取其中所有支持的文件",
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("FILE_LIST", "STRING")
+    RETURN_NAMES = ("文件列表", "文件信息")
+    FUNCTION = "load_file"
+    CATEGORY = "file/input"
+
+    def load_file(self, 单文件路径: str = "", 文件夹路径: str = "") -> Tuple[FileList, str]:
+        collected: List[Path] = []
+
+        # 1. 单文件路径（逗号分隔，支持多个）
+        if 单文件路径.strip():
+            for raw in 单文件路径.split(","):
+                p = Path(raw.strip().strip('"').strip("'"))
+                if not p.is_absolute():
+                    p = Path.cwd() / p
+                if not p.exists():
+                    raise ValueError(f"文件不存在: {p}")
+                if not p.is_file():
+                    raise ValueError(f"路径不是文件: {p}")
+                collected.append(p)
+
+        # 2. 文件夹路径
+        if 文件夹路径.strip():
+            folder = Path(文件夹路径.strip().strip('"').strip("'"))
+            if not folder.is_absolute():
+                folder = Path.cwd() / folder
+            if not folder.exists():
+                raise ValueError(f"文件夹不存在: {folder}")
+            if not folder.is_dir():
+                raise ValueError(f"路径不是文件夹: {folder}")
+            for p in sorted(folder.iterdir()):
+                if p.is_file() and p.suffix.lower() in DOCUMENT_MIME_TYPES:
+                    collected.append(p)
+            if not collected:
+                raise ValueError(f"文件夹中没有支持的文件: {folder}")
+
+        if not collected:
+            raise ValueError("请至少提供一个文件路径或文件夹路径")
+
+        # 去重（保持顺序）
+        seen = set()
+        unique: List[Path] = []
+        for p in collected:
+            key = str(p.resolve())
+            if key not in seen:
+                seen.add(key)
+                unique.append(p)
+
+        # 大小检查 & 读取
+        total_size = 0
+        file_list: FileList = []
+        info_lines = []
+
+        for p in unique:
+            ext = p.suffix.lower()
+            if ext not in DOCUMENT_MIME_TYPES:
+                print(f"LoadFile: 跳过不支持的文件类型 {p.name}")
+                continue
+
+            file_size = p.stat().st_size
+            if file_size > FILE_SIZE_LIMIT:
+                raise ValueError(
+                    f"文件 {p.name} 大小 {file_size / 1024 / 1024:.1f}MB 超过单文件 50MB 限制"
+                )
+            total_size += file_size
+            if total_size > TOTAL_FILE_SIZE_LIMIT:
+                raise ValueError(f"所有文件总大小超过 50MB 限制")
+
+            mime = DOCUMENT_MIME_TYPES[ext]
+            with open(p, "rb") as f:
+                b64 = base64.b64encode(f.read()).decode("utf-8")
+
+            file_list.append(FileData(
+                path=str(p),
+                filename=p.stem,
+                extension=ext,
+                mime_type=mime,
+                data=b64,
+                size=file_size,
+            ))
+            info_lines.append(f"  {p.name} ({file_size / 1024:.1f}KB, {mime})")
+            print(f"LoadFile: 加载 {p.name} ({file_size / 1024:.1f}KB)")
+
+        info = f"共 {len(file_list)} 个文件，总大小 {total_size / 1024:.1f}KB\n" + "\n".join(info_lines)
+        return (file_list, info)
@@ -0,0 +1,575 @@
+"""
+Nano Banana 节点 (V3)
+ComfyUI 自定义节点，用于调用异步生图模型
+使用 V3 DynamicCombo 实现模型-宽高比-分辨率动态联动
+"""
+
+import time
+import math
+import random
+import asyncio
+import aiohttp
+from concurrent.futures import ThreadPoolExecutor
+from typing import Callable, List, Optional
+
+import torch
+import numpy as np
+from PIL import Image
+
+from comfy_api.latest import io
+
+from ..utils.image_utils import tensor_to_pil, pil_to_tensor, parse_batch_prompts
+from ..utils.config import (
+    NETWORK_ROUTE_OPTIONS,
+    get_base_url_by_route,
+    get_api_key_or_raise,
+)
+from ..utils.nano_banana_async import generate_nano_banana_async
+from ..clients.gemini_client import GeminiAPIClient
+
+try:
+    from comfy.utils import ProgressBar
+    PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    PROGRESS_BAR_AVAILABLE = False
+
+try:
+    from comfy.model_management import processing_interrupted, InterruptProcessingException
+    INTERRUPT_AVAILABLE = True
+except ImportError:
+    INTERRUPT_AVAILABLE = False
+    InterruptProcessingException = RuntimeError
+    processing_interrupted = lambda: False
+
+REQUEST_LOG_ENABLED = False
+
+_NODE = "Nano Banana"
+_REQUEST_TIMEOUT = 900
+_INTERRUPT_CHECK_INTERVAL = 0.2
+
+_client_instance = None
+
+
+def _get_client():
+    global _client_instance
+    if _client_instance is None:
+        _client_instance = GeminiAPIClient()
+    return _client_instance
+
+
+async def _poll_interrupt():
+    while True:
+        await asyncio.sleep(_INTERRUPT_CHECK_INTERVAL)
+        if INTERRUPT_AVAILABLE and processing_interrupted():
+            return
+
+
+async def _run_with_interrupt(coro):
+    if not INTERRUPT_AVAILABLE:
+        return await coro
+
+    request_task = asyncio.ensure_future(coro)
+    interrupt_task = asyncio.ensure_future(_poll_interrupt())
+
+    done, pending = await asyncio.wait(
+        [request_task, interrupt_task],
+        return_when=asyncio.FIRST_COMPLETED,
+    )
+
+    for task in pending:
+        task.cancel()
+        try:
+            await task
+        except (asyncio.CancelledError, Exception):
+            pass
+
+    if interrupt_task in done and request_task not in done:
+        raise InterruptProcessingException()
+
+    return request_task.result()
+
+
+def _check_interrupt():
+    if INTERRUPT_AVAILABLE and processing_interrupted():
+        raise InterruptProcessingException()
+
+
+def _make_progress_callback(pbar) -> Optional[Callable[[float], None]]:
+    if pbar is None:
+        return None
+
+    last_progress = [0.0]
+
+    def _on_progress(progress: float) -> None:
+        try:
+            progress = max(0.0, min(float(progress), 1.0))
+        except (TypeError, ValueError):
+            return
+        if progress <= last_progress[0]:
+            return
+        pbar.update(progress - last_progress[0])
+        last_progress[0] = progress
+
+    return _on_progress
+
+
+def _images_to_tensor_safe(images: List[Image.Image], node_label: str) -> torch.Tensor:
+    if not images:
+        placeholder = Image.new('RGB', (512, 512), color=(128, 128, 128))
+        return pil_to_tensor([placeholder])
+
+    base_size = max(images, key=lambda img: img.size[0] * img.size[1]).size
+    matched = [img for img in images if img.size == base_size]
+    skipped = [img for img in images if img.size != base_size]
+
+    if skipped:
+        sizes_str = ", ".join(f"{img.size[0]}x{img.size[1]}" for img in skipped)
+        print(
+            f"{node_label}: 丢弃 {len(skipped)} 张较小尺寸的图 ({sizes_str})，"
+            f"仅输出最大尺寸 {base_size[0]}x{base_size[1]} 的 {len(matched)} 张"
+        )
+
+    return pil_to_tensor(matched)
+
+
+MODEL_ID_MAP = {
+    "Nano Banana Pro": "nano-banana-pro",
+    "Nano Banana 2": "nano-banana-2",
+    "Nano Banana": "nano-banana",
+}
+RESOLUTION_KEY_MAP = {
+    "512px": "0.5k",
+    "1K": "1k",
+    "2K": "2k",
+    "4K": "4k",
+}
+BILLING_SPECIAL_ONLY = {"nano-banana"}
+
+
+def _build_model_id(model_name: str, resolution: str, billing: str) -> str:
+    base = MODEL_ID_MAP.get(model_name, "nano-banana-pro")
+
+    if base == "nano-banana":
+        if billing == "官方":
+            raise ValueError(f"模型 \"{model_name}\" 仅支持特价计费")
+        return "nano-banana"
+
+    res_key = RESOLUTION_KEY_MAP.get(resolution, "2k")
+    is_official = (billing == "官方")
+
+    if base == "nano-banana-pro" and res_key == "1k" and not is_official:
+        return "nano-banana-pro"
+
+    if base == "nano-banana-2" and res_key == "0.5k":
+        if is_official:
+            raise ValueError("Nano Banana 2 的 512px 分辨率仅支持特价计费")
+        return "nano-banana-2-0.5k"
+
+    model_id = f"{base}-{res_key}"
+    if is_official:
+        model_id += "-official"
+    return model_id
+
+
+async def _generate_single(
+    session: aiohttp.ClientSession,
+    base_url: str,
+    api_key: str,
+    prompt: str,
+    model: str,
+    resolution: str,
+    aspect_ratio: str,
+    images: Optional[List[Image.Image]] = None,
+    enable_grounding: bool = False,
+    thinking_level: Optional[str] = None,
+    progress_callback: Optional[Callable[[float], None]] = None,
+) -> List[Image.Image]:
+    result_images, timing = await generate_nano_banana_async(
+        session=session,
+        base_url=base_url,
+        api_key=api_key,
+        prompt=prompt,
+        model=model,
+        resolution=resolution,
+        aspect_ratio=aspect_ratio,
+        images=images,
+        enable_grounding=enable_grounding,
+        thinking_level=thinking_level,
+        node_label="Nano Banana",
+        request_log_enabled=REQUEST_LOG_ENABLED,
+        check_interrupt=_check_interrupt,
+        progress_callback=progress_callback,
+    )
+    return result_images, timing["task_ms"], timing["parse_ms"]
+
+
+async def _generate_single_task(
+    session: aiohttp.ClientSession,
+    base_url: str,
+    api_key: str,
+    prompt: str,
+    model: str,
+    resolution: str,
+    aspect_ratio: str,
+    images: Optional[List[Image.Image]],
+    global_task_index: int,
+    enable_grounding: bool = False,
+    thinking_level: Optional[str] = None,
+    progress_callback: Optional[Callable[[float], None]] = None,
+) -> dict:
+    result = {
+        "global_task_index": global_task_index,
+        "prompt": prompt,
+        "success": False,
+        "generated_count": 0,
+        "output_images": [],
+        "error": None,
+    }
+    try:
+        gen_images, task_ms, parse_ms = await _generate_single(
+            session=session,
+            base_url=base_url,
+            api_key=api_key,
+            prompt=prompt,
+            model=model,
+            resolution=resolution,
+            aspect_ratio=aspect_ratio,
+            images=images if images else None,
+            enable_grounding=enable_grounding,
+            thinking_level=thinking_level,
+            progress_callback=progress_callback,
+        )
+        del task_ms, parse_ms
+        result["output_images"] = gen_images
+        result["success"] = True
+        result["generated_count"] = len(gen_images)
+    except InterruptProcessingException:
+        raise
+    except Exception as e:
+        result["error"] = str(e)
+    return result
+
+
+async def _process_batch_async(
+    base_url: str,
+    api_key: str,
+    prompts: List[str],
+    model: str,
+    resolution: str,
+    aspect_ratio: str,
+    images_per_prompt: int,
+    input_images: Optional[List[Image.Image]],
+    pbar=None,
+    enable_grounding: bool = False,
+    thinking_level: Optional[str] = None,
+) -> List[dict]:
+    tasks_def = []
+    for p_idx, prompt in enumerate(prompts):
+        for sub_idx in range(images_per_prompt):
+            tasks_def.append((p_idx, sub_idx, prompt))
+
+    total_tasks = len(tasks_def)
+    max_concurrent = 50
+    num_batches = math.ceil(total_tasks / max_concurrent)
+
+    all_results = []
+    completed = 0
+    success_count = 0
+    fail_count = 0
+
+    connector = aiohttp.TCPConnector(ssl=False, limit=0, limit_per_host=0)
+
+    async with aiohttp.ClientSession(connector=connector) as session:
+        for batch_idx in range(num_batches):
+            _check_interrupt()
+            start_idx = batch_idx * max_concurrent
+            end_idx = min(start_idx + max_concurrent, total_tasks)
+
+            tasks = []
+            for i in range(start_idx, end_idx):
+                _check_interrupt()
+                _, _, prompt = tasks_def[i]
+                task = asyncio.create_task(
+                    _generate_single_task(
+                        session=session,
+                        base_url=base_url,
+                        api_key=api_key,
+                        prompt=prompt,
+                        model=model,
+                        resolution=resolution,
+                        aspect_ratio=aspect_ratio,
+                        images=input_images,
+                        global_task_index=i,
+                        enable_grounding=enable_grounding,
+                        thinking_level=thinking_level,
+                        progress_callback=_make_progress_callback(pbar),
+                    )
+                )
+                tasks.append(task)
+
+            batch_results = []
+            for coro in asyncio.as_completed(tasks):
+                _check_interrupt()
+                result_data = None
+                try:
+                    result = await coro
+                    if isinstance(result, Exception):
+                        result_data = {"success": False, "error": str(result), "generated_count": 0, "output_images": [], "prompt": ""}
+                    else:
+                        result_data = result
+                except InterruptProcessingException:
+                    for task in tasks:
+                        task.cancel()
+                    await asyncio.gather(*tasks, return_exceptions=True)
+                    raise
+                except Exception as e:
+                    result_data = {"success": False, "error": str(e), "generated_count": 0, "output_images": [], "prompt": ""}
+
+                batch_results.append(result_data)
+                completed += 1
+                prompt_snippet = (result_data.get("prompt", "") or "")[:30]
+
+                if result_data and result_data.get("success", False):
+                    success_count += 1
+                    count = result_data.get("generated_count", 1)
+                    print(f"Nano Banana: [{completed}/{total_tasks}] {prompt_snippet}{'...' if len(prompt_snippet) >= 30 else ''} → ✓成功({count}张)")
+                else:
+                    fail_count += 1
+                    error_msg = result_data.get("error", "未知错误") if result_data else "未知错误"
+                    print(f"Nano Banana: [{completed}/{total_tasks}] {prompt_snippet}{'...' if len(prompt_snippet) >= 30 else ''} → ✗失败: {error_msg}")
+
+            all_results.extend(batch_results)
+            import gc; gc.collect()
+            await asyncio.sleep(0.1)
+
+    return all_results
+
+
+class NanoBanana(io.ComfyNode):
+
+    @classmethod
+    def define_schema(cls):
+        return io.Schema(
+            node_id="NanoBanana",
+            display_name="Nano Banana",
+            category="image/generation",
+            inputs=[
+                io.String.Input(
+                    "prompt",
+                    default="一个中国女子的OOTD",
+                    multiline=True,
+                ),
+                io.DynamicCombo.Input("模型", options=[
+                    io.DynamicCombo.Option("Nano Banana Pro", [
+                        io.Combo.Input("宽高比", options=[
+                            "智能", "1:1", "2:3", "3:2", "3:4", "4:3",
+                            "4:5", "5:4", "9:16", "16:9", "21:9",
+                        ], default="智能"),
+                        io.Combo.Input("分辨率", options=["1K", "2K", "4K"], default="2K"),
+                    ]),
+                    io.DynamicCombo.Option("Nano Banana 2", [
+                        io.Combo.Input("宽高比", options=[
+                            "智能", "1:1", "1:4", "1:8", "2:3", "3:2", "3:4",
+                            "4:1", "4:3", "4:5", "5:4", "8:1",
+                            "9:16", "16:9", "21:9",
+                        ], default="智能"),
+                        io.Combo.Input("分辨率", options=["512px", "1K", "2K", "4K"], default="2K"),
+                        io.Combo.Input("思考深度", options=["高", "低"], default="高"),
+                    ]),
+                    io.DynamicCombo.Option("Nano Banana", [
+                        io.Combo.Input("宽高比", options=[
+                            "智能", "1:1", "2:3", "3:2", "3:4", "4:3",
+                            "4:5", "5:4", "9:16", "16:9", "21:9",
+                        ], default="智能"),
+                        io.Combo.Input("分辨率", options=["1K"], default="1K"),
+                    ]),
+                ]),
+                io.Int.Input("生图数量", default=1, min=1, max=1000, step=1),
+                io.Combo.Input("网络", options=NETWORK_ROUTE_OPTIONS, default="全球加速"),
+                io.Combo.Input("计费", options=["特价", "官方"], default="特价"),
+                io.Combo.Input("谷歌搜索", options=["关闭", "打开"], default="关闭"),
+                io.Int.Input("seed", default=0, min=0, max=0xFFFFFFFFFFFFFFFF),
+                io.Image.Input("参考图1", optional=True),
+                io.Image.Input("参考图2", optional=True),
+                io.Image.Input("参考图3", optional=True),
+                io.Image.Input("参考图4", optional=True),
+                io.Image.Input("参考图5", optional=True),
+                io.Image.Input("参考图6", optional=True),
+                io.Image.Input("参考图7", optional=True),
+                io.Image.Input("参考图8", optional=True),
+                io.Image.Input("参考图9", optional=True),
+            ],
+            outputs=[
+                io.Image.Output(display_name="输出图像"),
+            ],
+        )
+
+    @classmethod
+    def execute(cls, prompt, 模型, 生图数量, 计费, 网络, 谷歌搜索, seed, **kwargs) -> io.NodeOutput:
+        start_time = time.time()
+        was_interrupted = False
+
+        model_name = 模型["模型"]
+        宽高比 = 模型["宽高比"]
+        分辨率 = 模型["分辨率"]
+        思考深度 = 模型.get("思考深度")
+
+        enable_grounding = (谷歌搜索 == "打开")
+
+        thinking_level = None
+        if model_name == "Nano Banana 2" and 思考深度:
+            thinking_level = "High" if 思考深度 == "高" else "Low"
+
+        actual_model = _build_model_id(model_name, 分辨率, 计费)
+
+        api_key = get_api_key_or_raise("O1KEY_API_KEY")
+        base_url = get_base_url_by_route(网络)
+
+        pbar = ProgressBar(生图数量) if PROGRESS_BAR_AVAILABLE else None
+
+        try:
+            random.seed(seed)
+            np.random.seed(seed % (2**32))
+
+            input_images = []
+            for i in range(1, 10):
+                key = f"参考图{i}"
+                if key in kwargs and kwargs[key] is not None:
+                    pil_imgs = tensor_to_pil(kwargs[key])
+                    input_images.extend(pil_imgs)
+
+            if len(input_images) > 14:
+                raise ValueError(f"输入图像数量 {len(input_images)} 超过限制 14 张")
+
+            batch_prompts = parse_batch_prompts(prompt)
+
+            grounding_str = " | 谷歌搜索接地" if enable_grounding else ""
+            thinking_str = f" | 思考:{thinking_level}" if thinking_level else ""
+            if batch_prompts:
+                num_prompts = len(batch_prompts)
+                total_images = num_prompts * 生图数量
+                mode_str = f"批量提示词模式 ({num_prompts}个提示词)"
+                if input_images:
+                    mode_str += f" (输入{len(input_images)}张)"
+                print(f"Nano Banana: {mode_str} | {分辨率} {宽高比} | 共{total_images}张{grounding_str}{thinking_str}")
+            else:
+                mode_str = f"图生图模式 (输入{len(input_images)}张)" if input_images else "文生图模式"
+                print(f"Nano Banana: {mode_str} | {分辨率} {宽高比} | {生图数量}张{grounding_str}{thinking_str}")
+
+            if batch_prompts or 生图数量 > 1:
+                prompts = batch_prompts if batch_prompts else [prompt]
+                images_per_prompt = 生图数量
+                total_tasks = len(prompts) * images_per_prompt
+
+                if pbar is not None:
+                    pbar = ProgressBar(total_tasks)
+
+                def run_async_in_thread():
+                    loop = asyncio.new_event_loop()
+                    asyncio.set_event_loop(loop)
+                    try:
+                        return loop.run_until_complete(
+                            _run_with_interrupt(_process_batch_async(
+                                base_url=base_url,
+                                api_key=api_key,
+                                prompts=prompts,
+                                model=actual_model,
+                                resolution=分辨率,
+                                aspect_ratio=宽高比,
+                                images_per_prompt=images_per_prompt,
+                                input_images=input_images,
+                                pbar=pbar,
+                                enable_grounding=enable_grounding,
+                                thinking_level=thinking_level,
+                            ))
+                        )
+                    finally:
+                        loop.close()
+
+                with ThreadPoolExecutor(max_workers=1) as executor:
+                    future = executor.submit(run_async_in_thread)
+                    try:
+                        results = future.result(timeout=_REQUEST_TIMEOUT)
+                    except TimeoutError:
+                        raise RuntimeError(f"任务执行超时（{_REQUEST_TIMEOUT}秒）")
+
+                success_count = sum(1 for r in results if r.get("success", False))
+                fail_count = len(results) - success_count
+                elapsed = time.time() - start_time
+                time_str = f"{elapsed:.3f}s" if elapsed < 1 else f"{elapsed:.2f}s"
+                print(f"完成！总耗时 {time_str} | 成功: {success_count}/{total_tasks} | 失败: {fail_count}")
+
+                output_images = []
+                for r in results:
+                    output_images.extend(r.get("output_images", []))
+                if not output_images:
+                    placeholder = Image.new('RGB', (512, 512), color=(128, 128, 128))
+                    output_images = [placeholder]
+
+                output_tensor = _images_to_tensor_safe(output_images, _NODE)
+                import gc; gc.collect()
+                return io.NodeOutput(output_tensor)
+
+            else:
+                def run_single():
+                    loop = asyncio.new_event_loop()
+                    asyncio.set_event_loop(loop)
+                    try:
+                        async def _do():
+                            connector = aiohttp.TCPConnector(ssl=False)
+                            async with aiohttp.ClientSession(connector=connector) as session:
+                                return await _generate_single(
+                                    session=session,
+                                    base_url=base_url,
+                                    api_key=api_key,
+                                    prompt=prompt,
+                                    model=actual_model,
+                                    resolution=分辨率,
+                                    aspect_ratio=宽高比,
+                                    images=input_images if input_images else None,
+                                    enable_grounding=enable_grounding,
+                                    thinking_level=thinking_level,
+                                    progress_callback=_make_progress_callback(pbar),
+                                )
+                        return loop.run_until_complete(_run_with_interrupt(_do()))
+                    finally:
+                        loop.close()
+
+                with ThreadPoolExecutor(max_workers=1) as executor:
+                    future = executor.submit(run_single)
+                    generated_images, task_ms, parse_ms = future.result(timeout=_REQUEST_TIMEOUT)
+
+                output_tensor = _images_to_tensor_safe(generated_images, _NODE)
+                elapsed = time.time() - start_time
+                time_str = f"{elapsed:.3f}s" if elapsed < 1 else f"{elapsed:.2f}s"
+                task_str = f"{task_ms/1000:.2f}s"
+                parse_str = f"{parse_ms/1000:.2f}s"
+                print(f"完成！总耗时 {time_str} | 异步任务 {task_str} | 解析 {parse_str} | 成功 {len(generated_images)}张")
+
+                import gc; gc.collect()
+                return io.NodeOutput(output_tensor)
+
+        except InterruptProcessingException:
+            was_interrupted = True
+            print("Nano Banana: 用户取消")
+            raise
+        except ValueError as e:
+            if str(e) == "未授权！":
+                print("请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            raise ValueError(str(e)) from None
+        except RuntimeError as e:
+            raise RuntimeError(str(e)) from None
+        except Exception as e:
+            raise RuntimeError(str(e)) from None
+        finally:
+            if not was_interrupted:
+                try:
+                    client = _get_client()
+                    client.base_url = base_url
+                    balance_data = client.query_balance_sync()
+                    balance_info = client.format_balance_info(balance_data)
+                    print(f"Nano Banana: {balance_info}")
+                except Exception:
+                    pass
+            import gc; gc.collect()
@@ -1,381 +0,0 @@
-"""
-Nano Banana Pro 节点
-ComfyUI 自定义节点，用于调用 Gemini 3 Pro 模型生成图像
-"""
-
-import time
-import random
-from typing import Optional, Tuple
-
-import torch
-import numpy as np
-from PIL import Image
-
-from ..utils.image_utils import tensor_to_pil, pil_to_tensor, parse_batch_prompts
-from ..clients.gemini_client import GeminiAPIClient
-from ..models_config import get_enabled_models, get_model_description
-
-# 导入 ComfyUI 原生进度条
-try:
-    from comfy.utils import ProgressBar
-    PROGRESS_BAR_AVAILABLE = True
-except ImportError:
-    PROGRESS_BAR_AVAILABLE = False
-    print("⚠️ NanoBananaPro: comfy.utils.ProgressBar 不可用，将只使用终端进度显示")
-
-
-class NanoBananaPro:
-    """
-    Nano Banana Pro 节点
-    
-    功能：
-    - 文生图：基于提示词生成图像
-    - 图生图：基于输入图像和提示词生成新图像
-    - 批量生成：支持并发生成多张图像
-    
-    注意：
-    - 支持的模型列表从 models_config.py 动态加载
-    - 要添加/禁用模型，请编辑 models_config.py 文件
-    """
-    
-    # 支持的模型列表（从配置文件动态加载）
-    MODELS = None  # 将在 INPUT_TYPES 中动态获取
-    
-    # 支持的宽高比列表
-    ASPECT_RATIOS = [
-        "1:1", "4:3", "3:4", "16:9", "9:16",
-        "2:3", "3:2", "4:5", "5:4", "21:9"
-    ]
-    
-    # 支持的分辨率列表
-    RESOLUTIONS = ["1K", "2K", "4K"]
-    
-    def __init__(self):
-        """初始化节点"""
-        self.client = None
-    
-    @classmethod
-    def INPUT_TYPES(cls):
-        """
-        定义输入参数
-        
-        ComfyUI 节点规范：
-        - required: 必选参数
-        - optional: 可选参数
-        """
-        # 从配置文件动态获取启用的模型列表
-        enabled_models = get_enabled_models()
-        
-        # 如果没有启用的模型，使用空列表（会导致节点不可用，提示用户配置）
-        if not enabled_models:
-            enabled_models = ["请在 models_config.py 中启用至少一个模型"]
-        
-        # 创建9个独立的图像输入
-        optional_inputs = {}
-        for i in range(1, 10):  # 1-9
-            optional_inputs[f"参考图{i}"] = ("IMAGE",)
-        
-        return {
-            "required": {
-                "prompt": ("STRING", {
-                    "default": "一个中国女子的OOTD",
-                    "multiline": True
-                }),
-                "模型": (enabled_models, {
-                    "default": enabled_models[0]
-                }),
-                "宽高比": (cls.ASPECT_RATIOS, {
-                    "default": "1:1"
-                }),
-                "分辨率": (cls.RESOLUTIONS, {
-                    "default": "2K"
-                }),
-                "生图数量": ("INT", {
-                    "default": 1,
-                    "min": 1,
-                    "max": 1000,
-                    "step": 1
-                }),
-                "像素缩放": ("BOOLEAN", {
-                    "default": False
-                }),
-                "分辨率像素": ("FLOAT", {
-                    "default": 1.0,
-                    "min": 0.1,
-                    "max": 100.0,
-                    "step": 0.1,
-                    "display": "number"
-                }),
-                "seed": ("INT", {
-                    "default": 0,
-                    "min": 0,
-                    "max": 0xffffffffffffffff
-                })
-            },
-            "optional": optional_inputs
-        }
-    
-    # 返回值类型
-    RETURN_TYPES = ("IMAGE",)
-    RETURN_NAMES = ("输出图像",)
-    
-    # 执行函数名
-    FUNCTION = "generate"
-    
-    # 节点分类
-    CATEGORY = "image/generation"
-    
-    def resize_to_megapixels(
-        self,
-        image: Image.Image,
-        target_megapixels: float
-    ) -> Image.Image:
-        """
-        将图像缩放到指定的总像素数，保持纵横比
-        
-        Args:
-            image: PIL Image 对象
-            target_megapixels: 目标像素数（百万像素）
-        
-        Returns:
-            缩放后的 PIL Image
-        
-        Example:
-            >>> resized = self.resize_to_megapixels(img, 2.0)  # 缩放到2百万像素
-        """
-        # 计算当前像素数
-        current_pixels = image.width * image.height
-        target_pixels = int(target_megapixels * 1_000_000)
-        
-        # 如果当前像素数已经接近目标，则不缩放
-        if abs(current_pixels - target_pixels) / target_pixels < 0.05:
-            return image
-        
-        # 计算缩放比例
-        scale = (target_pixels / current_pixels) ** 0.5
-        
-        # 计算新尺寸
-        new_width = int(image.width * scale)
-        new_height = int(image.height * scale)
-        
-        # 确保至少为1像素
-        new_width = max(1, new_width)
-        new_height = max(1, new_height)
-        
-        # 使用 Lanczos 重采样
-        resized_image = image.resize((new_width, new_height), Image.Resampling.LANCZOS)
-        
-        return resized_image
-    
-    def validate_inputs(
-        self,
-        images: Optional[torch.Tensor],
-        batch_size: int
-    ) -> None:
-        """
-        验证输入参数
-        
-        Args:
-            images: 输入图像张量（可选）
-            batch_size: 批次大小
-        
-        Raises:
-            ValueError: 如果输入参数不合法
-        """
-        # 检查图像数量
-        if images is not None:
-            num_images = images.shape[0]
-            if num_images > 14:
-                raise ValueError(
-                    f"输入图像数量 {num_images} 超过限制 14 张，请减少输入图像数量"
-                )
-        
-        # 检查批次大小
-        if batch_size < 1 or batch_size > 1000:
-            raise ValueError(
-                f"批次大小 {batch_size} 超出范围 [1, 1000]"
-            )
-    
-    def generate(
-        self,
-        prompt: str,
-        模型: str,
-        宽高比: str,
-        分辨率: str,
-        生图数量: int,
-        像素缩放: bool,
-        分辨率像素: float,
-        seed: int,
-        **kwargs
-    ) -> Tuple[torch.Tensor]:
-        """
-        生成图像
-        
-        Args:
-            prompt: 提示词
-            模型: 模型名称
-            宽高比: 宽高比
-            分辨率: 分辨率
-            生图数量: 批次大小
-            像素缩放: 是否启用像素缩放
-            分辨率像素: 目标像素数（百万像素）
-            seed: 随机种子
-            **kwargs: 动态参考图输入 (参考图1-9)
-        
-        Returns:
-            生成的图像张量 (IMAGE,)
-        """
-        start_time = time.time()
-        
-        # 创建 ComfyUI 原生进度条
-        pbar = None
-        if PROGRESS_BAR_AVAILABLE:
-            pbar = ProgressBar(生图数量)
-        
-        try:
-            # 设置随机种子（用于本地随机操作）
-            random.seed(seed)
-            np.random.seed(seed % (2**32))
-            
-            # 初始化 API 客户端
-            if self.client is None:
-                try:
-                    self.client = GeminiAPIClient()
-                except ValueError as e:
-                    raise ValueError(f"初始化失败: {str(e)}")
-            
-            # 收集独立输入的参考图
-            input_images = []
-            for i in range(1, 10):  # 1-9
-                key = f"参考图{i}"
-                if key in kwargs and kwargs[key] is not None:
-                    pil_imgs = tensor_to_pil(kwargs[key])
-                    input_images.extend(pil_imgs)
-            
-            # 验证输入图像数量
-            if input_images:
-                if len(input_images) > 14:
-                    raise ValueError(
-                        f"输入图像数量 {len(input_images)} 超过限制 14 张，请减少输入图像数量"
-                    )
-            
-            # 应用像素缩放（如果启用）
-            if input_images and 像素缩放:
-                scaled_images = []
-                for img in input_images:
-                    scaled = self.resize_to_megapixels(img, 分辨率像素)
-                    scaled_images.append(scaled)
-                input_images = scaled_images
-                print(f"Nano Banana Pro: 已缩放 {len(scaled_images)} 张图像到 {分辨率像素}M 像素")
-            
-            # 转换为 API 所需的格式
-            if input_images:
-                print(f"Nano Banana Pro: 图生图模式 (输入 {len(input_images)} 张图像)")
-            
-            # 解析批量提示词
-            batch_prompts = parse_batch_prompts(prompt)
-            
-            # 统计变量
-            success_count = 0
-            fail_count = 0
-            
-            # 进度回调 - 实时显示每个任务的完成状态，并更新 ComfyUI 进度条
-            def progress_callback(current, total, success, error_msg=None):
-                nonlocal success_count, fail_count
-                if success:
-                    success_count += 1
-                    print(f"Nano Banana Pro: ✓ [{current}/{total}] 第 {success_count} 张生成成功")
-                else:
-                    fail_count += 1
-                    error_brief = error_msg[:50] + "..." if error_msg and len(error_msg) > 50 else error_msg
-                    print(f"Nano Banana Pro: ✗ [{current}/{total}] 生成失败 - {error_brief}")
-                
-                # 更新 ComfyUI 原生进度条
-                if pbar is not None:
-                    pbar.update(1)
-            
-            # 根据是否有批量提示词选择生成模式
-            if batch_prompts:
-                # 批量提示词模式
-                num_prompts = len(batch_prompts)
-                total_images = num_prompts * 生图数量
-                print(f"Nano Banana Pro: 批量提示词模式 ({num_prompts} 个提示词 × {生图数量} 张/提示词 = {total_images} 张图)")
-                print(f"Nano Banana Pro: 发送请求")
-                print(f"Nano Banana Pro: 生图中...")
-                
-                # 重新创建进度条以匹配实际总数
-                if pbar is not None:
-                    pbar = ProgressBar(total_images)
-                
-                generated_images = self.client.generate_multi_prompts_sync(
-                    prompts=batch_prompts,
-                    model=模型,
-                    resolution=分辨率,
-                    aspect_ratio=宽高比,
-                    images_per_prompt=生图数量,
-                    images=input_images,
-                    progress_callback=progress_callback
-                )
-                
-                if fail_count > 0:
-                    print(f"Nano Banana Pro: 生图完成 (成功: {success_count}, 失败: {fail_count})")
-                else:
-                    print(f"Nano Banana Pro: 全部生图成功！")
-            else:
-                # 单提示词模式
-                print(f"Nano Banana Pro: {'图生图' if input_images else '文生图'}模式")
-                print(f"Nano Banana Pro: 发送请求")
-                print(f"Nano Banana Pro: 生图中...")
-                
-                generated_images = self.client.generate_sync(
-                    prompt=prompt,
-                    model=模型,
-                    resolution=分辨率,
-                    aspect_ratio=宽高比,
-                    batch_size=生图数量,
-                    images=input_images,
-                    progress_callback=progress_callback
-                )
-                
-                if fail_count > 0:
-                    print(f"Nano Banana Pro: 生图完成 (成功: {success_count}, 失败: {fail_count})")
-                else:
-                    print(f"Nano Banana Pro: 全部生图成功！")
-            
-            # 转换输出图像
-            output_tensor = pil_to_tensor(generated_images)
-            
-            # 计算耗时
-            elapsed = time.time() - start_time
-            print(f"Nano Banana Pro: 完成生图 (耗时: {elapsed:.2f}s, 成功生成 {len(generated_images)} 张图像)")
-            
-            return (output_tensor,)
-        
-        except ValueError as e:
-            # 检测是否为授权错误
-            if str(e) == "未授权！":
-                print("请联系作者授权后方可使用！")
-            else:
-                # 用户输入错误
-                print(f"Nano Banana Pro: 输入错误 - {str(e)}")
-            raise
-        
-        except RuntimeError as e:
-            # API 或网络错误
-            print(f"Nano Banana Pro: API 错误 - {str(e)}")
-            raise
-        
-        except Exception as e:
-            # 其他未知错误
-            print(f"Nano Banana Pro: 未知错误 - {str(e)}")
-            raise
-        
-        finally:
-            # 无论成功或失败，都尝试查询余额
-            if self.client is not None:
-                try:
-                    balance_data = self.client.query_balance_sync()
-                    balance_info = self.client.format_balance_info(balance_data)
-                    print(f"Nano Banana Pro: {balance_info}")
-                except Exception as e:
-                    print(f"Nano Banana Pro: ⚠️ 余额查询失败 - {str(e)}")
@@ -0,0 +1,254 @@
+"""
+Single-node new-api Veo 3.1 generator.
+
+The node submits a /v1/videos task, waits for completion, downloads the mp4,
+and returns ComfyUI's native VIDEO object for the built-in Save Video node.
+"""
+
+import os
+from io import BytesIO
+from typing import Optional, Tuple
+
+from ..clients.newapi_veo_client import NewAPIVeoClient
+from ..utils.image_utils import tensor_to_pil
+from ..utils.config import NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+
+try:
+    import folder_paths
+    FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    FOLDER_PATHS_AVAILABLE = False
+
+try:
+    from comfy.utils import ProgressBar
+    PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    ProgressBar = None
+    PROGRESS_BAR_AVAILABLE = False
+
+try:
+    from comfy_api.input_impl import VideoFromFile
+except Exception:
+    VideoFromFile = None
+
+
+MODEL_OPTIONS = [
+    "veo-3.1",
+]
+
+DURATION_OPTIONS = ["4", "6", "8"]
+ASPECT_RATIO_OPTIONS = ["16:9", "9:16"]
+RESOLUTION_OPTIONS = ["720p", "1080p"]
+
+TARGET_SIZE_MAP = {
+    ("720p", "16:9"): (1280, 720),
+    ("720p", "9:16"): (720, 1280),
+    ("1080p", "16:9"): (1920, 1080),
+    ("1080p", "9:16"): (1080, 1920),
+}
+
+
+def _get_output_dir() -> str:
+    if FOLDER_PATHS_AVAILABLE:
+        return folder_paths.get_output_directory()
+
+    plugin_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+    comfy_root = os.path.dirname(os.path.dirname(plugin_dir))
+    return os.path.join(comfy_root, "output")
+
+
+def _get_download_dir() -> str:
+    output_dir = _get_output_dir()
+    video_dir = os.path.join(output_dir, "newapi_veo")
+    os.makedirs(video_dir, exist_ok=True)
+    return video_dir
+
+
+def _fit_image_to_target(image, target_size: Tuple[int, int]):
+    from PIL import Image as PILImage
+
+    target_w, target_h = target_size
+    src_w, src_h = image.size
+    src_ratio = src_w / src_h
+    target_ratio = target_w / target_h
+
+    if src_w == target_w and src_h == target_h:
+        return image
+
+    resample = PILImage.Resampling.LANCZOS if hasattr(PILImage, "Resampling") else PILImage.LANCZOS
+
+    if src_ratio > target_ratio:
+        scale = target_h / src_h
+        new_w = round(src_w * scale)
+        image = image.resize((new_w, target_h), resample=resample)
+        left = max(0, (new_w - target_w) // 2)
+        image = image.crop((left, 0, left + target_w, target_h))
+    else:
+        scale = target_w / src_w
+        new_h = round(src_h * scale)
+        image = image.resize((target_w, new_h), resample=resample)
+        top = max(0, (new_h - target_h) // 2)
+        image = image.crop((0, top, target_w, top + target_h))
+
+    return image
+
+
+def _image_to_png_bytes(image_tensor, resolution: str, aspect_ratio: str) -> Optional[bytes]:
+    if image_tensor is None:
+        return None
+
+    pil_images = tensor_to_pil(image_tensor)
+    if not pil_images:
+        return None
+
+    image = pil_images[0]
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+
+    target_size = TARGET_SIZE_MAP.get((resolution, aspect_ratio))
+    if target_size is not None:
+        original_size = image.size
+        image = _fit_image_to_target(image, target_size)
+        if image.size != original_size:
+            print(
+                "NewAPI Veo: input image fitted "
+                f"{original_size[0]}x{original_size[1]} -> {image.size[0]}x{image.size[1]}"
+            )
+
+    buffer = BytesIO()
+    image.save(buffer, format="PNG")
+    image_bytes = buffer.getvalue()
+    print(
+        "NewAPI Veo: input_reference PNG "
+        f"{len(image_bytes) / 1024:.0f} KB ({image.size[0]}x{image.size[1]})"
+    )
+    return image_bytes
+
+
+class Google31Video:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词": (
+                    "STRING",
+                    {
+                        "default": "A cinematic shot of a small robot walking through a rainy neon street.",
+                        "multiline": True,
+                    },
+                ),
+                "负向提示词": ("STRING", {"default": "", "multiline": True}),
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型": (MODEL_OPTIONS, {"default": MODEL_OPTIONS[0]}),
+                "时长": (DURATION_OPTIONS, {"default": "8"}),
+                "宽高比": (ASPECT_RATIO_OPTIONS, {"default": "16:9"}),
+                "分辨率": (RESOLUTION_OPTIONS, {"default": "1080p"}),
+                "生成音频": (["打开", "关闭"], {"default": "打开"}),
+                "seed": (
+                    "INT",
+                    {
+                        "default": -1,
+                        "min": -1,
+                        "max": 0xFFFFFFFFFFFFFFFF,
+                        "step": 1,
+                    },
+                ),
+            },
+            "optional": {
+                "参考图像": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO",)
+    RETURN_NAMES = ("视频",)
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Video"
+
+    DESCRIPTION = (
+        "Submit a new-api /v1/videos Veo 3.1 task, poll until complete, "
+        "download the mp4, and output native VIDEO for ComfyUI Save Video."
+    )
+
+    def generate(
+        self,
+        提示词: str,
+        负向提示词: str,
+        网络线路: str,
+        模型: str,
+        时长: str,
+        宽高比: str,
+        分辨率: str,
+        生成音频: str,
+        seed: int,
+        参考图像=None,
+    ):
+        if VideoFromFile is None:
+            raise RuntimeError("当前 ComfyUI 版本不支持原生 VIDEO 输入实现 VideoFromFile。")
+
+        prompt = (提示词 or "").strip()
+        if not prompt:
+            raise ValueError("提示词不能为空。")
+
+        duration_value = int(时长)
+        if duration_value not in (4, 6, 8):
+            raise ValueError("时长仅支持 4、6、8。")
+        if 宽高比 not in ASPECT_RATIO_OPTIONS:
+            raise ValueError("宽高比仅支持 16:9 或 9:16。")
+        if 分辨率 not in RESOLUTION_OPTIONS:
+            raise ValueError("分辨率仅支持 720p 或 1080p。")
+
+        output_dir = _get_download_dir()
+        image_bytes = _image_to_png_bytes(参考图像, 分辨率, 宽高比)
+
+        pbar = ProgressBar(100) if PROGRESS_BAR_AVAILABLE else None
+        last_progress = [0]
+        last_status = [""]
+
+        def progress_callback(progress: int, status: str, elapsed: float):
+            if status != last_status[0]:
+                print(
+                    "NewAPI Veo: polling "
+                    f"status={status} | elapsed={elapsed:.0f}s"
+                )
+                last_status[0] = status
+
+            progress = max(0, min(100, int(progress or 0)))
+            if pbar is not None and progress > last_progress[0]:
+                pbar.update(progress - last_progress[0])
+                last_progress[0] = progress
+
+        client = NewAPIVeoClient(base_url=get_base_url_by_route(网络线路))
+
+        result = client.generate_video_sync(
+            prompt=prompt,
+            model=模型,
+            duration=duration_value,
+            aspect_ratio=宽高比,
+            resolution=分辨率,
+            output_dir=output_dir,
+            negative_prompt=负向提示词,
+            generate_audio=(生成音频 == "打开"),
+            image_bytes=image_bytes,
+            poll_interval=10,
+            timeout=900,
+            progress_callback=progress_callback,
+        )
+
+        video_path = result["video_path"]
+        video = VideoFromFile(video_path)
+
+        print(
+            "NewAPI Veo: completed "
+            f"| task_id={result['task_id']} | video={video_path}"
+        )
+
+        return (video,)
+
+
+NODE_CLASS_MAPPINGS = {
+    "Google31Video": Google31Video,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "Google31Video": "Google 3.1 Video",
+}
@@ -0,0 +1,36 @@
+"""
+o1key 去背景节点
+基于 rembg 实现，支持 CPU 推理
+"""
+
+import numpy as np
+import torch
+
+
+class O1keyRemoveBackground:
+    """
+    移除图像背景，输出 RGBA 透明图层
+
+    基于 rembg (ISNet-General-Use) 模型，支持 CPU 推理。
+    首次运行会自动下载模型（约 170MB）。
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "image": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("IMAGE",)
+    RETURN_NAMES = ("RGBA图像",)
+    FUNCTION = "remove_bg"
+    CATEGORY = "o1key/image"
+
+    def remove_bg(self, image):
+        from ..utils.rembg_utils import remove_background_tensor
+        print("[o1key 去背景] 正在处理...")
+        result = remove_background_tensor(image)
+        print(f"[o1key 去背景] 完成，输出 {result.shape[0]} 张 RGBA")
+        return (result,)
@@ -0,0 +1,188 @@
+"""
+图像元数据去除节点
+
+提供批量去除已有图片中元数据的功能
+"""
+
+import os
+
+from PIL import Image
+from PIL.PngImagePlugin import PngInfo
+
+# 支持的图片格式
+SUPPORTED_EXTENSIONS = {'.png', '.jpg', '.jpeg', '.webp', '.bmp', '.tiff', '.tif'}
+
+
+def _save_image_clean(image: Image.Image, path: str, fmt: str = None, quality: int = 95) -> None:
+    """
+    保存图像，不包含任何元数据
+
+    通过提取纯像素数据并重建全新的 Image 对象，确保没有任何元数据残留。
+
+    Args:
+        image: PIL Image 对象
+        path: 保存路径
+        fmt: 图像格式（PNG/JPEG/WEBP），为 None 时根据扩展名推断
+        quality: JPEG/WEBP 质量（1-100）
+    """
+    # 确保 RGB 模式
+    if image.mode != 'RGB':
+        image = image.convert('RGB')
+
+    # 提取纯像素数据，重建全新的 Image 对象
+    # 使用 tobytes() + frombytes() 确保只保留像素数据，彻底断开与原图像的关联
+    pixel_data = image.tobytes()
+    clean = Image.frombytes('RGB', image.size, pixel_data)
+
+    # 显式清空 info 字典，确保不会有任何残留元数据
+    clean.info = {}
+
+    # 推断格式
+    if fmt is None:
+        ext = os.path.splitext(path)[1].lower()
+        format_map = {
+            '.png': 'PNG',
+            '.jpg': 'JPEG',
+            '.jpeg': 'JPEG',
+            '.webp': 'WEBP',
+            '.bmp': 'BMP',
+            '.tiff': 'TIFF',
+            '.tif': 'TIFF',
+        }
+        fmt = format_map.get(ext, 'PNG')
+
+    # 构建保存参数（确保不写入任何元数据）
+    save_kwargs = {}
+    if fmt == 'PNG':
+        save_kwargs['pnginfo'] = PngInfo()  # 空的 PngInfo，不包含任何文本块
+    elif fmt == 'JPEG':
+        save_kwargs['quality'] = quality
+        # 不传 exif 参数，自然不会写入 EXIF 数据
+    elif fmt == 'WEBP':
+        save_kwargs['quality'] = quality
+        save_kwargs['exif'] = b""  # 显式清空 EXIF
+
+    clean.save(path, format=fmt, **save_kwargs)
+
+
+# ============================================================================
+# 批量去除元数据
+# ============================================================================
+
+class BatchCleanMetadata:
+    """
+    批量去除文件夹中图片元数据的节点
+
+    功能：
+    - 指定文件夹路径，批量处理其中所有图片
+    - 去除 EXIF、PNG tEXt 块、ComfyUI 工作流等所有元数据
+    - 支持保存到原目录（添加 _nometa 后缀）或覆盖原文件
+    - 支持 PNG/JPG/JPEG/WEBP/BMP/TIFF 格式
+
+    使用场景：
+    - 已经保存了一批含有 AI 元数据的图片，需要批量清理
+    - 批量处理指定文件夹中的所有图片
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        """
+        定义输入参数
+
+        Returns:
+            输入参数配置字典
+        """
+        return {
+            "required": {
+                "文件夹路径": ("STRING", {"default": ""}),
+                "覆盖原文件": ("BOOLEAN", {"default": False}),
+            }
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("处理结果",)
+    OUTPUT_NODE = True
+    FUNCTION = "batch_clean"
+    CATEGORY = "image"
+
+    DESCRIPTION = (
+        "批量去除文件夹中图片的元数据。\n"
+        "支持 PNG/JPG/JPEG/WEBP/BMP/TIFF 格式。\n"
+        "默认在原文件名后添加 _nometa 后缀保存，也可选择覆盖原文件。"
+    )
+
+    def batch_clean(
+        self,
+        文件夹路径: str,
+        覆盖原文件: bool = False,
+    ) -> tuple:
+        """
+        批量去除文件夹中图片的元数据
+
+        Args:
+            文件夹路径: 待处理图片所在的文件夹路径
+            覆盖原文件: 是否覆盖原文件（False 则添加 _nometa 后缀）
+
+        Returns:
+            处理结果字符串
+
+        Raises:
+            ValueError: 文件夹路径无效
+        """
+        if not 文件夹路径 or not 文件夹路径.strip():
+            raise ValueError("请输入文件夹路径")
+
+        folder = 文件夹路径.strip()
+
+        if not os.path.isdir(folder):
+            raise ValueError(f"文件夹路径无效或不存在: {folder}")
+
+        # 扫描支持的图片文件
+        files = []
+        for f in sorted(os.listdir(folder)):
+            ext = os.path.splitext(f)[1].lower()
+            if ext in SUPPORTED_EXTENSIONS:
+                files.append(f)
+
+        if not files:
+            msg = f"文件夹中未找到支持的图片文件 ({', '.join(SUPPORTED_EXTENSIONS)})"
+            print(f"批量去除元数据: {msg}")
+            return (msg,)
+
+        print(f"批量去除元数据: 找到 {len(files)} 张图片，开始处理...")
+
+        success_count = 0
+        fail_count = 0
+
+        for f in files:
+            try:
+                src_path = os.path.join(folder, f)
+                img = Image.open(src_path)
+
+                if 覆盖原文件:
+                    dst_path = src_path
+                else:
+                    name, ext = os.path.splitext(f)
+                    dst_path = os.path.join(folder, f"{name}_nometa{ext}")
+
+                _save_image_clean(img, dst_path)
+                success_count += 1
+
+            except Exception as e:
+                print(f"批量去除元数据: 处理 {f} 失败 - {str(e)}")
+                fail_count += 1
+
+        # 构建结果消息
+        if fail_count > 0:
+            msg = f"处理完成: 成功 {success_count} 张, 失败 {fail_count} 张"
+        else:
+            msg = f"处理完成: 全部 {success_count} 张成功"
+
+        if not 覆盖原文件:
+            msg += " (已添加 _nometa 后缀)"
+        else:
+            msg += " (已覆盖原文件)"
+
+        print(f"批量去除元数据: {msg}")
+
+        return (msg,)
@@ -0,0 +1,94 @@
+"""保存图像节点 - 支持 PNG/JPEG/WebP 格式输出"""
+
+import os
+import json
+import numpy as np
+from PIL import Image
+from PIL.PngImagePlugin import PngInfo
+
+import folder_paths
+from comfy.cli_args import args
+
+
+class SaveImageFormat:
+    """保存图像，支持 PNG / JPEG / WebP 三种格式"""
+
+    FORMATS = ["PNG", "JPEG", "WebP"]
+
+    def __init__(self):
+        self.output_dir = folder_paths.get_output_directory()
+        self.type = "output"
+        self.compress_level = 4
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "图像": ("IMAGE",),
+                "文件名前缀": ("STRING", {"default": "ComfyUI"}),
+                "输出格式": (cls.FORMATS, {"default": "PNG"}),
+            },
+            "optional": {},
+            "hidden": {
+                "prompt": "PROMPT",
+                "extra_pnginfo": "EXTRA_PNGINFO",
+            },
+        }
+
+    RETURN_TYPES = ()
+    FUNCTION = "save_images"
+    OUTPUT_NODE = True
+    CATEGORY = "image"
+    DESCRIPTION = "保存图像，支持 PNG / JPEG / WebP 格式输出。"
+
+    _EXT_MAP = {"PNG": ".png", "JPEG": ".jpg", "WebP": ".webp"}
+
+    def save_images(self, 图像=None, 文件名前缀="ComfyUI", 输出格式="PNG",
+                    prompt=None, extra_pnginfo=None):
+        images = 图像
+        filename_prefix = 文件名前缀
+        format = 输出格式
+        full_output_folder, filename, counter, subfolder, filename_prefix = \
+            folder_paths.get_save_image_path(
+                filename_prefix, self.output_dir,
+                images[0].shape[1], images[0].shape[0]
+            )
+
+        ext = self._EXT_MAP.get(format, ".png")
+        results = []
+
+        for batch_number, image in enumerate(images):
+            i = 255.0 * image.cpu().numpy()
+            img = Image.fromarray(np.clip(i, 0, 255).astype(np.uint8))
+
+            filename_with_batch_num = filename.replace("%batch_num%", str(batch_number))
+            file = f"{filename_with_batch_num}_{counter:05}_{ext}"
+
+            filepath = os.path.join(full_output_folder, file)
+
+            if format == "PNG":
+                metadata = None
+                if not args.disable_metadata:
+                    metadata = PngInfo()
+                    if prompt is not None:
+                        metadata.add_text("prompt", json.dumps(prompt))
+                    if extra_pnginfo is not None:
+                        for x in extra_pnginfo:
+                            metadata.add_text(x, json.dumps(extra_pnginfo[x]))
+                img.save(filepath, pnginfo=metadata,
+                         compress_level=self.compress_level)
+            elif format == "JPEG":
+                if img.mode == "RGBA":
+                    img = img.convert("RGB")
+                img.save(filepath, quality=100, optimize=True)
+            elif format == "WebP":
+                img.save(filepath, lossless=True)
+
+            results.append({
+                "filename": file,
+                "subfolder": subfolder,
+                "type": self.type,
+            })
+            counter += 1
+
+        return {"ui": {"images": results}}
@@ -0,0 +1,237 @@
+"""
+o1key SavePSD 节点
+将多个 IMAGE 图层合成为分层 PSD 文件
+手写 PSD 二进制格式，零外部依赖（仅 numpy + Pillow）
+"""
+
+import os
+import struct
+import time
+import numpy as np
+import torch
+from PIL import Image
+
+import folder_paths
+
+
+def _pad_even(data: bytes) -> bytes:
+    if len(data) % 2:
+        return data + b"\x00"
+    return data
+
+
+def _pad4(data: bytes) -> bytes:
+    return data + (b"\x00" * ((4 - (len(data) % 4)) % 4))
+
+
+def _pascal_name(name: str) -> bytes:
+    raw = name.encode("macroman", errors="replace")[:255]
+    data = bytes([len(raw)]) + raw
+    return _pad4(data)
+
+
+def _unicode_name_block(name: str) -> bytes:
+    payload = struct.pack(">I", len(name)) + name.encode("utf-16be")
+    block = b"8BIM" + b"luni" + struct.pack(">I", len(payload)) + _pad_even(payload)
+    return block
+
+
+def _layer_extra_data(name: str) -> bytes:
+    data = b""
+    data += struct.pack(">I", 0)  # layer mask data length
+    data += struct.pack(">I", 0)  # layer blending ranges length
+    data += _pascal_name(name)
+    data += _unicode_name_block(name)
+    return data
+
+
+def _alpha_bbox(rgba_arr: np.ndarray):
+    """找到 RGBA 数组中非透明区域的 bounding box。"""
+    alpha = rgba_arr[:, :, 3]
+    rows = np.any(alpha > 0, axis=1)
+    cols = np.any(alpha > 0, axis=0)
+    if not rows.any():
+        return None
+    top = int(np.argmax(rows))
+    bottom = int(len(rows) - np.argmax(rows[::-1]))
+    left = int(np.argmax(cols))
+    right = int(len(cols) - np.argmax(cols[::-1]))
+    return top, left, bottom, right
+
+
+def write_psd(filepath: str, layers: list, canvas_w: int, canvas_h: int):
+    """
+    写入 PSD 文件。
+
+    layers: [(name, rgba_array), ...] 从底到顶排列
+    rgba_array: numpy uint8 [H, W, 4]
+    """
+    records = []
+    channel_data_blocks = []
+    layers_top_to_bottom = list(reversed(layers))
+
+    for name, rgba in layers_top_to_bottom:
+        bbox = _alpha_bbox(rgba)
+        if not bbox:
+            continue
+        top, left, bottom, right = bbox
+        cropped = rgba[top:bottom, left:right]
+
+        # PLACEHOLDER_CHANNELS
+
+        channels = [
+            (0, cropped[:, :, 0].tobytes(order="C")),
+            (1, cropped[:, :, 1].tobytes(order="C")),
+            (2, cropped[:, :, 2].tobytes(order="C")),
+            (-1, cropped[:, :, 3].tobytes(order="C")),
+        ]
+        channel_info = b""
+        data_block = b""
+        for channel_id, data in channels:
+            channel_info += struct.pack(">hI", channel_id, 2 + len(data))
+            data_block += struct.pack(">H", 0) + data  # raw compression
+
+        extra = _layer_extra_data(name)
+        record = b""
+        record += struct.pack(">iiii", top, left, bottom, right)
+        record += struct.pack(">H", len(channels))
+        record += channel_info
+        record += b"8BIM" + b"norm"
+        record += bytes([255, 0, 0, 0])  # opacity=255, clipping, flags, filler
+        record += struct.pack(">I", len(extra)) + extra
+        records.append(record)
+        channel_data_blocks.append(data_block)
+
+    if not records:
+        raise ValueError("所有图层均为空（完全透明），无法生成 PSD")
+
+    # Layer and Mask Information
+    layer_info = struct.pack(">h", len(records))
+    layer_info += b"".join(records) + b"".join(channel_data_blocks)
+    layer_info = _pad_even(layer_info)
+    layer_info_block = struct.pack(">I", len(layer_info)) + layer_info
+    global_mask = struct.pack(">I", 0)
+    layer_mask_payload = layer_info_block + global_mask
+    layer_and_mask = struct.pack(">I", len(layer_mask_payload)) + layer_mask_payload
+
+    # PLACEHOLDER_COMPOSITE
+
+    # Composite preview (flattened image for compatibility)
+    comp = Image.new("RGBA", (canvas_w, canvas_h), (255, 255, 255, 255))
+    for name, rgba in layers:
+        layer_img = Image.fromarray(rgba, "RGBA")
+        comp.alpha_composite(layer_img)
+    comp_rgb = np.asarray(comp.convert("RGB"), dtype=np.uint8)
+    composite_data = (
+        struct.pack(">H", 0)
+        + comp_rgb[:, :, 0].tobytes(order="C")
+        + comp_rgb[:, :, 1].tobytes(order="C")
+        + comp_rgb[:, :, 2].tobytes(order="C")
+    )
+
+    # Write PSD file
+    with open(filepath, "wb") as f:
+        # Header
+        f.write(b"8BPS")
+        f.write(struct.pack(">H", 1))  # version
+        f.write(b"\x00" * 6)  # reserved
+        f.write(struct.pack(">HIIHH", 3, canvas_h, canvas_w, 8, 3))
+        # Color Mode Data
+        f.write(struct.pack(">I", 0))
+        # Image Resources
+        f.write(struct.pack(">I", 0))
+        # Layer and Mask
+        f.write(layer_and_mask)
+        # Composite Image Data
+        f.write(composite_data)
+
+
+# PLACEHOLDER_NODE
+
+class O1keySavePSD:
+    """
+    将多个 IMAGE 输入合成为分层 PSD 文件
+
+    每个输入作为独立图层，支持 RGBA 透明通道。
+    图层从下到上排列（图层1在最底部）。
+    使用 bbox 裁剪优化文件大小，包含合成预览层。
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "批次图像": ("IMAGE", {
+                    "tooltip": "批次图像输入，每张图自动作为独立图层（支持RGBA透明）",
+                }),
+            },
+            "optional": {
+                "图层名称": ("STRING", {
+                    "default": "",
+                    "multiline": True,
+                    "tooltip": "每行一个图层名称，与图层顺序对应。留空则自动命名。",
+                }),
+                "文件名前缀": ("STRING", {
+                    "default": "o1key_layers",
+                    "tooltip": "输出 PSD 文件名前缀",
+                }),
+            },
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("文件路径",)
+    FUNCTION = "save_psd"
+    CATEGORY = "o1key/image"
+    OUTPUT_NODE = True
+
+    def save_psd(self, 批次图像, 图层名称: str = "", 文件名前缀: str = "o1key_layers", **kwargs):
+        # 将批次 tensor [B, H, W, C] 拆为单张列表
+        if 批次图像.dim() == 3:
+            layer_tensors = [批次图像]
+        else:
+            layer_tensors = [批次图像[i] for i in range(批次图像.shape[0])]
+
+        names = [n.strip() for n in 图层名称.split("\n") if n.strip()]
+
+        # 确定画布尺寸
+        max_h, max_w = 0, 0
+        for t in layer_tensors:
+            h, w = t.shape[0], t.shape[1]
+            max_h = max(max_h, h)
+            max_w = max(max_w, w)
+
+        # 转换为 [(name, rgba_array), ...] 格式
+        layers = []
+        for idx, tensor in enumerate(layer_tensors):
+            arr = (tensor.cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
+            h, w = arr.shape[0], arr.shape[1]
+            channels = arr.shape[2] if arr.ndim == 3 else 1
+
+            if channels == 3:
+                rgba = np.zeros((max_h, max_w, 4), dtype=np.uint8)
+                rgba[:h, :w, :3] = arr
+                rgba[:h, :w, 3] = 255
+            elif channels == 4:
+                rgba = np.zeros((max_h, max_w, 4), dtype=np.uint8)
+                rgba[:h, :w] = arr
+            else:
+                rgba = np.zeros((max_h, max_w, 4), dtype=np.uint8)
+                rgba[:h, :w, 0] = rgba[:h, :w, 1] = rgba[:h, :w, 2] = arr[:, :, 0] if arr.ndim == 3 else arr
+                rgba[:h, :w, 3] = 255
+
+            name = names[idx] if idx < len(names) else f"图层 {idx + 1}"
+            layers.append((name, rgba))
+            print(f"[o1key SavePSD] 图层 '{name}': {w}×{h}")
+
+        # 写入 PSD
+        output_dir = folder_paths.get_output_directory()
+        timestamp = time.strftime("%Y%m%d_%H%M%S")
+        filename = f"{文件名前缀}_{timestamp}.psd"
+        filepath = os.path.join(output_dir, filename)
+
+        write_psd(filepath, layers, max_w, max_h)
+
+        size_kb = os.path.getsize(filepath) / 1024
+        print(f"[o1key SavePSD] 完成: {filepath} ({size_kb:.0f}KB, "
+              f"{len(layers)} 层, {max_w}×{max_h})")
+        return (filepath,)
@@ -0,0 +1,490 @@
+"""
+Seedance 视频生成节点
+节点列表:
+  - Seedance: 文生视频 / 图生视频 / 首尾帧生视频（根据图片输入自动切换模式）
+"""
+
+import base64
+import io
+import json
+import os
+import tempfile
+
+import aiohttp
+import torch
+
+from ..clients.seedance_client import SeedanceClient
+from ..clients.gemini_client import GeminiAPIClient
+from ..utils.image_utils import tensor_to_pil, pil_to_tensor
+from ..utils.r2_uploader import upload_video, upload_audio
+from ..utils.config import NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+
+from comfy_api.latest import InputImpl
+
+
+# ── 模型列表 ──────────────────────────────────────────────────────────────────
+
+_MODELS = [
+    "doubao-seedance-2-0-260128",
+]
+
+_RESOLUTIONS = ["720p", "1080p", "480p"]
+
+_MAX_IMAGE_BYTES = 30 * 1024 * 1024
+_MAX_REQUEST_BODY_BYTES = 64 * 1024 * 1024
+
+
+# ── 模型能力判断 ──────────────────────────────────────────────────────────────
+
+def _supports_camera_fixed(model: str) -> bool:
+    """2.0 系列不支持固定镜头"""
+    return False  # 当前仅 2.0 模型，均不支持
+
+
+# ── 工具函数 ──────────────────────────────────────────────────────────────────
+
+def _format_mb(size_bytes: int) -> str:
+    return f"{size_bytes / 1024 / 1024:.2f}MB"
+
+
+def _tensor_to_base64_url(tensor, label: str = "图片") -> str:
+    """ComfyUI IMAGE tensor → data:image/png;base64,xxx"""
+    pil_images = tensor_to_pil(tensor)
+    image = pil_images[0]
+    if image.mode == "RGBA":
+        image = image.convert("RGB")
+
+    buffered = io.BytesIO()
+    image.save(buffered, format="PNG")
+    image_bytes = buffered.getvalue()
+    image_size = len(image_bytes)
+
+    if image_size > _MAX_IMAGE_BYTES:
+        raise ValueError(
+            f"Seedance {label}大小 {_format_mb(image_size)} 超过单张图片 "
+            f"{_format_mb(_MAX_IMAGE_BYTES)} 限制，请先压缩或缩小图片。"
+        )
+
+    b64 = base64.b64encode(image_bytes).decode("utf-8")
+    return f"data:image/png;base64,{b64}"
+
+
+def _validate_request_body_size(body: dict, tag: str):
+    body_size = len(json.dumps(body, ensure_ascii=False).encode("utf-8"))
+    if body_size > _MAX_REQUEST_BODY_BYTES:
+        raise ValueError(
+            f"{tag} 请求体大小 {_format_mb(body_size)} 超过 "
+            f"{_format_mb(_MAX_REQUEST_BODY_BYTES)} 限制，请减少参考图片数量或降低图片尺寸。"
+        )
+    print(
+        f"[{tag}] 请求体大小: {_format_mb(body_size)} "
+        f"(限制 {_format_mb(_MAX_REQUEST_BODY_BYTES)})"
+    )
+
+
+
+async def _url_to_tensor(url: str) -> torch.Tensor:
+    """从 URL 下载图片并转为 ComfyUI IMAGE tensor，失败时返回 None"""
+    try:
+        from PIL import Image
+        async with aiohttp.ClientSession() as session:
+            async with session.get(url, allow_redirects=True) as resp:
+                if resp.status != 200:
+                    return None
+                data = await resp.read()
+        img = Image.open(io.BytesIO(data)).convert("RGB")
+        return pil_to_tensor([img])
+    except Exception as e:
+        print(f"[Seedance] 末帧图片下载失败: {e}")
+        return None
+
+
+def _show_balance():
+    """完成后打印余额（静默失败）"""
+    try:
+        client = GeminiAPIClient()
+        data = client.query_balance_sync()
+        print(f"Seedance: {client.format_balance_info(data)}")
+    except Exception:
+        pass
+
+
+def _make_pbar():
+    try:
+        from comfy.utils import ProgressBar
+        return ProgressBar(100)
+    except Exception:
+        return None
+
+
+def _make_callbacks(tag: str, pbar):
+    def on_stage(stage: str):
+        if stage == "submitting":
+            print(f"[{tag}] 提交中...")
+            if pbar: pbar.update_absolute(0, 100)
+        elif stage.startswith("submitted:"):
+            print(f"[{tag}] 已提交 → {stage.split(':', 1)[1]}")
+            if pbar: pbar.update_absolute(5, 100)
+        elif stage == "downloading":
+            print(f"[{tag}] 下载视频中...")
+            if pbar: pbar.update_absolute(99, 100)
+        elif stage == "done":
+            print(f"[{tag}] 完成")
+            if pbar: pbar.update_absolute(100, 100)
+
+    def on_progress(pct: int):
+        if pbar: pbar.update_absolute(5 + int(pct * 0.94), 100)
+
+    return on_stage, on_progress
+
+
+
+# ── 统一节点 ─────────────────────────────────────────────────────────────────
+#
+#   模式由图片输入自动判断：
+#     首帧 = None              → T2V 文生视频   （联网搜索生效）
+#     首帧 = 图片，尾帧 = None → I2V 图生视频   （固定镜头生效，当前 2.0 不支持故忽略）
+#     首帧 = 图片，尾帧 = 图片 → FlipFlop 首尾帧（联网搜索/固定镜头均忽略）
+
+class Seedance:
+    """Seedance 视频生成（文生视频 / 图生视频 / 首尾帧，自动判断模式）"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词":           ("STRING", {"multiline": True, "default": ""}),
+                "网络线路":         (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型":             (_MODELS, {"default": "doubao-seedance-2-0-260128"}),
+                "分辨率":           (_RESOLUTIONS, {"default": "720p"}),
+                "宽高比":           (["16:9", "adaptive", "9:16", "1:1", "4:3", "3:4", "21:9"],
+                                     {"default": "16:9"}),
+                "时长秒(-1=自动)":  ("INT", {"default": 5, "min": -1, "max": 30, "step": 1}),
+                "生成音频":         (["关闭", "打开"], {"default": "关闭"}),
+                "联网搜索":         (["关闭", "打开"], {"default": "关闭"}),
+                "返回末帧图片":     (["关闭", "打开"], {"default": "关闭"}),
+                "seed":             ("INT", {"default": 0, "min": 0, "max": 0xffffffff}),
+            },
+            "optional": {
+                "首帧图片":         ("IMAGE",),
+                "尾帧图片":         ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO", "IMAGE")
+    RETURN_NAMES = ("视频", "末帧图片")
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Seedance"
+
+    async def generate(self, **kwargs):
+        prompt      = kwargs["提示词"].strip()
+        model       = kwargs["模型"]
+        resolution  = kwargs["分辨率"]
+        ratio       = kwargs["宽高比"]
+        duration    = kwargs["时长秒(-1=自动)"]
+        gen_audio   = kwargs["生成音频"] == "打开"
+        web_search  = kwargs["联网搜索"] == "打开"
+        return_last = kwargs["返回末帧图片"] == "打开"
+        seed        = kwargs.get("seed", 0)
+        first_image = kwargs.get("首帧图片", None)
+        last_image  = kwargs.get("尾帧图片", None)
+
+        # 模式判断
+        if first_image is None and last_image is not None:
+            raise ValueError("请同时接入首帧图片，或仅接入首帧图片。")
+        if first_image is None:
+            mode = "t2v"
+            tag  = "Seedance文生视频"
+            file_prefix = "seedance_t2v"
+        elif last_image is None:
+            mode = "i2v"
+            tag  = "Seedance图生视频"
+            file_prefix = "seedance_i2v"
+        else:
+            mode = "flipflop"
+            tag  = "Seedance首尾帧"
+            file_prefix = "seedance_flip"
+
+        if not prompt:
+            raise ValueError("提示词不能为空。")
+        if duration == -1 and mode == "t2v":
+            pass  # 2.0 均支持自动时长
+        elif duration == -1 and mode != "t2v":
+            pass  # 2.0 均支持自动时长
+
+        metadata: dict = {
+            "resolution": resolution,
+            "watermark":  False,
+        }
+        if ratio != "adaptive":
+            metadata["ratio"] = ratio
+        if duration != -1:
+            metadata["duration"] = duration
+        if gen_audio:
+            metadata["generate_audio"] = True
+        if return_last:
+            metadata["return_last_frame"] = True
+        if seed != 0:
+            metadata["seed"] = seed
+
+        # 模式专属参数
+        if mode == "t2v":
+            if web_search:
+                metadata["tools"] = [{"type": "web_search"}]
+            body = {
+                "model":    model,
+                "prompt":   prompt,
+                "metadata": metadata,
+            }
+
+        elif mode == "i2v":
+            first_url = _tensor_to_base64_url(first_image, "首帧图片")
+            metadata["content"] = [
+                {
+                    "type":      "image_url",
+                    "image_url": {"url": first_url},
+                    "role":      "first_frame",
+                },
+                {"type": "text", "text": prompt},
+            ]
+            body = {
+                "model":    model,
+                "prompt":   prompt,
+                "images":   [first_url],
+                "metadata": metadata,
+            }
+
+        else:  # flipflop
+            first_url = _tensor_to_base64_url(first_image, "首帧图片")
+            last_url  = _tensor_to_base64_url(last_image, "尾帧图片")
+            metadata["content"] = [
+                {
+                    "type":      "image_url",
+                    "image_url": {"url": first_url},
+                    "role":      "first_frame",
+                },
+                {
+                    "type":      "image_url",
+                    "image_url": {"url": last_url},
+                    "role":      "last_frame",
+                },
+                {"type": "text", "text": prompt},
+            ]
+            body = {
+                "model":    model,
+                "prompt":   prompt,
+                "images":   [first_url],
+                "metadata": metadata,
+            }
+
+        _validate_request_body_size(body, tag)
+
+        # 保存路径（临时文件，避免与下游保存节点重复落盘）
+        _, save_path = tempfile.mkstemp(suffix=".mp4", prefix=f"{file_prefix}_")
+
+        client            = SeedanceClient()
+        client.base_url   = get_base_url_by_route(kwargs.get("网络线路", "全球加速"))
+        pbar              = _make_pbar()
+        on_stage, on_prog = _make_callbacks(tag, pbar)
+
+        try:
+            result_path, last_frame_url = await client.generate_async(
+                body=body, save_path=save_path,
+                on_stage=on_stage, on_progress=on_prog,
+            )
+            last_frame_tensor = None
+            if return_last and last_frame_url:
+                last_frame_tensor = await _url_to_tensor(last_frame_url)
+            return (InputImpl.VideoFromFile(result_path), last_frame_tensor)
+        finally:
+            _show_balance()
+
+
+# ── 多模态参考生视频节点 ──────────────────────────────────────────────────────
+
+class SeedanceMultiModal:
+    """Seedance 2.0 多模态参考生视频（参考图片 + 参考视频 + 参考音频 + 文本）"""
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "提示词":           ("STRING", {"multiline": True, "default": ""}),
+                "网络线路":         (NETWORK_ROUTE_OPTIONS, {"default": "全球加速"}),
+                "模型":             (_MODELS, {"default": "doubao-seedance-2-0-260128"}),
+                "分辨率":           (_RESOLUTIONS, {"default": "720p"}),
+                "宽高比":           (["adaptive", "16:9", "9:16", "1:1", "4:3", "3:4", "21:9"],
+                                     {"default": "adaptive"}),
+                "时长秒(-1=自动)":  ("INT", {"default": 5, "min": -1, "max": 15, "step": 1}),
+                "生成音频":         (["关闭", "打开"], {"default": "关闭"}),
+                "联网搜索":         (["关闭", "打开"], {"default": "关闭"}),
+                "返回末帧图片":     (["关闭", "打开"], {"default": "关闭"}),
+                "seed":             ("INT", {"default": 0, "min": 0, "max": 0xffffffff}),
+            },
+            "optional": {
+                "参考图片":         ("IMAGE",),
+                "参考视频1":        ("VIDEO",),
+                "参考视频2":        ("VIDEO",),
+                "参考视频3":        ("VIDEO",),
+                "参考音频1":        ("AUDIO",),
+                "参考音频2":        ("AUDIO",),
+                "参考音频3":        ("AUDIO",),
+            },
+        }
+
+    RETURN_TYPES = ("VIDEO", "IMAGE")
+    RETURN_NAMES = ("视频", "末帧图片")
+    FUNCTION = "generate"
+    CATEGORY = "comfyui_o1key/Seedance"
+    INPUT_IS_LIST = True
+
+    async def generate(self, **kwargs):
+        # INPUT_IS_LIST=True 时所有参数都是列表，取第一个元素
+        def _first(v, default=None):
+            if isinstance(v, list):
+                return v[0] if v else default
+            return v if v is not None else default
+
+        prompt      = _first(kwargs.get("提示词"), "").strip()
+        model       = _first(kwargs.get("模型"))
+        resolution  = _first(kwargs.get("分辨率"))
+        ratio       = _first(kwargs.get("宽高比"))
+        duration    = _first(kwargs.get("时长秒(-1=自动)"), 5)
+        gen_audio   = _first(kwargs.get("生成音频"), "关闭") == "打开"
+        web_search  = _first(kwargs.get("联网搜索"), "关闭") == "打开"
+        return_last = _first(kwargs.get("返回末帧图片"), "关闭") == "打开"
+        seed        = _first(kwargs.get("seed"), 0)
+        network_route = _first(kwargs.get("网络线路"), "全球加速")
+
+        # 参考图片：INPUT_IS_LIST 时是 [tensor, tensor, ...] 列表，直接保留
+        raw_images = kwargs.get("参考图片", None)
+        ref_images = [img for img in raw_images if img is not None] if raw_images else None
+
+        ref_videos  = [_first(kwargs.get(f"参考视频{i}")) for i in range(1, 4)]
+        ref_audios  = [_first(kwargs.get(f"参考音频{i}")) for i in range(1, 4)]
+
+        ref_videos  = [v for v in ref_videos if v is not None]
+        ref_audios  = [a for a in ref_audios if a is not None]
+
+        # ── 校验 ──────────────────────────────────────────────────────────
+        has_image = bool(ref_images)
+        has_video = len(ref_videos) > 0
+        has_audio = len(ref_audios) > 0
+
+        if not has_image and not has_video and not has_audio and not prompt:
+            raise ValueError("至少需要提供参考图片、参考视频或提示词之一。")
+        if has_audio and not has_image and not has_video:
+            raise ValueError("不可单独输入音频，请至少连接一张参考图片或一个参考视频。")
+
+        # ── 构建 content 列表 ─────────────────────────────────────────────
+        content = []
+
+        # 参考图片（批次，最多9张）
+        if has_image:
+            imgs = ref_images[:9]
+            if len(ref_images) > 9:
+                print(f"[SeedanceMultiModal] 参考图片超过9张，仅取前9张（共{len(ref_images)}张）")
+            for idx, img_tensor in enumerate(imgs, start=1):
+                # 每个 tensor 可能是 [1,H,W,C] 或 [H,W,C]，统一确保有 batch 维
+                if img_tensor.dim() == 3:
+                    img_tensor = img_tensor.unsqueeze(0)
+                url = _tensor_to_base64_url(img_tensor, f"参考图片{idx}")
+                content.append({
+                    "type":      "image_url",
+                    "image_url": {"url": url},
+                    "role":      "reference_image",
+                })
+
+        # 参考视频（最多3个）
+        for v in ref_videos:
+            url = await upload_video(v)
+            content.append({
+                "type":      "video_url",
+                "video_url": {"url": url},
+                "role":      "reference_video",
+            })
+
+        # 参考音频（最多3段）
+        for a in ref_audios:
+            url = await upload_audio(a)
+            content.append({
+                "type":      "audio_url",
+                "audio_url": {"url": url},
+                "role":      "reference_audio",
+            })
+
+        # 文本提示词（放最后）
+        if prompt:
+            content.append({"type": "text", "text": prompt})
+
+        if not content:
+            raise ValueError("content 为空，请至少提供参考图片、参考视频或提示词。")
+
+        # ── 构建请求体（new-api 兼容格式）──────────────────────────────────
+        metadata: dict = {
+            "resolution": resolution,
+            "watermark":  False,
+            "content":    content,
+        }
+
+        if ratio != "adaptive":
+            metadata["ratio"] = ratio
+        if duration != -1:
+            metadata["duration"] = duration
+        if gen_audio:
+            metadata["generate_audio"] = True
+        if return_last:
+            metadata["return_last_frame"] = True
+        if seed != 0:
+            metadata["seed"] = seed
+        if web_search:
+            metadata["tools"] = [{"type": "web_search"}]
+
+        # 顶层 image：取第一张参考图的 base64（new-api 单图字段）
+        first_image_url = next(
+            (item["image_url"]["url"] for item in content if item["type"] == "image_url"),
+            None,
+        )
+
+        body = {
+            "model":    model,
+            "prompt":   prompt if prompt else " ",
+            "metadata": metadata,
+        }
+        if first_image_url:
+            body["image"] = first_image_url
+
+        _validate_request_body_size(body, "Seedance多模态")
+
+        # ── 保存路径（临时文件，避免与下游保存节点重复落盘）──────────────────
+        _, save_path = tempfile.mkstemp(suffix=".mp4", prefix="seedance_mm_")
+
+        client            = SeedanceClient()
+        client.base_url   = get_base_url_by_route(network_route)
+        pbar              = _make_pbar()
+        on_stage, on_prog = _make_callbacks("Seedance多模态", pbar)
+
+        try:
+            result_path, last_frame_url = await client.generate_async(
+                body=body, save_path=save_path,
+                on_stage=on_stage, on_progress=on_prog,
+            )
+            last_frame_tensor = None
+            if return_last and last_frame_url:
+                last_frame_tensor = await _url_to_tensor(last_frame_url)
+            return (InputImpl.VideoFromFile(result_path), last_frame_tensor)
+        finally:
+            _show_balance()
+
+
+# ── 节点注册 ──────────────────────────────────────────────────────────────────
+
+NODE_CLASS_MAPPINGS = {
+    "Seedance":          Seedance,
+    "SeedanceMultiModal": SeedanceMultiModal,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "Seedance":          "Seedance 视频生成",
+    "SeedanceMultiModal": "Seedance 多模态参考生视频",
+}
@@ -0,0 +1,526 @@
+"""
+Sora 视频生成节点
+ComfyUI 自定义节点，调用 Sora API 生成视频
+"""
+
+import os
+import re
+import time
+from math import gcd
+from typing import Optional, Tuple
+
+import torch
+
+from ..utils.image_utils import tensor_to_pil
+from ..clients.sora_client import SoraClient
+from ..models_config import (
+    get_enabled_sora_models,
+    get_all_sora_seconds,
+    get_all_sora_sizes,
+    get_sora_supported_seconds,
+    get_sora_supported_sizes,
+    get_sora_seconds_with_labels,
+    get_sora_sizes_with_labels,
+    SORA_MODELS,
+)
+
+try:
+    import folder_paths
+    FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    FOLDER_PATHS_AVAILABLE = False
+
+try:
+    from comfy.utils import ProgressBar
+    PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    PROGRESS_BAR_AVAILABLE = False
+    print("⚠️ SoraVideo: comfy.utils.ProgressBar 不可用，将只使用终端进度显示")
+
+
+def _size_to_display(size: str) -> str:
+    """
+    将 'WxH' 格式的分辨率转换为友好显示名。
+
+    例如：
+        "720x1280"  → "720P  9:16"
+        "1280x720"  → "720P 16:9"
+        "1024x1792" → "1K  4:7"
+        "1792x1024" → "1K  7:4"
+
+    Args:
+        size: 分辨率字符串，格式 "WxH"
+
+    Returns:
+        友好显示名字符串
+    """
+    parts = size.lower().split("x")
+    w, h = int(parts[0]), int(parts[1])
+    short_side = min(w, h)
+    if short_side >= 3840:
+        res = "4K"
+    elif short_side >= 1920:
+        res = "2K"
+    elif short_side >= 1080:
+        res = "1K"
+    elif short_side >= 720:
+        res = "720P"
+    elif short_side >= 480:
+        res = "480P"
+    else:
+        res = f"{short_side}P"
+    g = gcd(w, h)
+    ratio = f"{w // g}:{h // g}"
+    return f"{res} {ratio} ({size})"
+
+
+def _build_size_display_map(sizes: list) -> dict:
+    """
+    构建 显示名 → 实际值 映射字典。
+
+    Args:
+        sizes: 实际分辨率列表，如 ["720x1280", "1280x720"]
+
+    Returns:
+        字典，key 为显示名，value 为实际分辨率字符串
+    """
+    mapping = {}
+    for size in sizes:
+        display = _size_to_display(size)
+        if display in mapping:
+            # 极少数情况下防止重名
+            display = f"{display} ({size})"
+        mapping[display] = size
+    return mapping
+
+
+def _get_video_output_dir() -> str:
+    """获取视频输出目录: ComfyUI/output/video"""
+    if FOLDER_PATHS_AVAILABLE:
+        base = folder_paths.get_output_directory()
+    else:
+        plugin_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+        base = os.path.join(os.path.dirname(os.path.dirname(plugin_dir)), "output")
+    video_dir = os.path.join(base, "video")
+    os.makedirs(video_dir, exist_ok=True)
+    return video_dir
+
+
+def _get_next_counter(directory: str, prefix: str) -> int:
+    """扫描目录，获取下一个可用的文件计数器"""
+    if not os.path.exists(directory):
+        return 1
+    pattern = re.compile(rf"^{re.escape(prefix)}_(\d+)")
+    max_counter = 0
+    for f in os.listdir(directory):
+        m = pattern.match(f)
+        if m:
+            max_counter = max(max_counter, int(m.group(1)))
+    return max_counter + 1
+
+
+def _fit_image_to_target(image, target_size: str):
+    """
+    将参考图片按 "等比缩放覆盖 + 居中裁剪" 策略适配到目标分辨率。
+
+    策略 (Cover Crop)：
+    1. 比较图片宽高比和目标宽高比
+    2. 等比缩放，使图片最短边刚好覆盖目标对应边（图片完全覆盖目标区域）
+    3. 居中裁剪多余部分，得到精确目标尺寸
+
+    Args:
+        image: PIL Image 对象
+        target_size: 目标分辨率字符串，格式 "WxH"（如 "720x1280"）
+
+    Returns:
+        适配后的 PIL Image 对象
+    """
+    from PIL import Image as PILImage
+
+    # 解析目标尺寸
+    parts = target_size.lower().split("x")
+    target_w, target_h = int(parts[0]), int(parts[1])
+
+    src_w, src_h = image.size
+    src_ratio = src_w / src_h
+    target_ratio = target_w / target_h
+
+    # 宽高比一致且尺寸不超过目标，无需处理
+    if abs(src_ratio - target_ratio) < 0.01 and src_w <= target_w and src_h <= target_h:
+        return image
+
+    print(f"Sora: 参考图片 {src_w}x{src_h} (比例 {src_ratio:.2f}) → 目标 {target_w}x{target_h} (比例 {target_ratio:.2f})")
+
+    # 获取高质量重采样滤波器
+    resample = PILImage.Resampling.LANCZOS if hasattr(PILImage, "Resampling") else PILImage.LANCZOS
+
+    # Cover Crop: 缩放使图片完全覆盖目标区域，然后居中裁剪
+    if src_ratio > target_ratio:
+        # 图片更宽：以高度为基准缩放，裁左右
+        scale = target_h / src_h
+        new_w = round(src_w * scale)
+        new_h = target_h
+        image = image.resize((new_w, new_h), resample=resample)
+        # 居中裁剪宽度
+        left = (new_w - target_w) // 2
+        image = image.crop((left, 0, left + target_w, target_h))
+    else:
+        # 图片更高（或一样）：以宽度为基准缩放，裁上下
+        scale = target_w / src_w
+        new_w = target_w
+        new_h = round(src_h * scale)
+        image = image.resize((new_w, new_h), resample=resample)
+        # 居中裁剪高度
+        top = (new_h - target_h) // 2
+        image = image.crop((0, top, target_w, top + target_h))
+
+    print(f"Sora: 参考图片已适配为 {image.size[0]}x{image.size[1]}")
+    return image
+
+
+def _compress_image_for_upload(
+    image,
+    target_size: Optional[str] = None,
+) -> bytes:
+    """
+    将 PIL Image 适配目标分辨率并编码为 PNG 字节，用于上传。
+
+    ============================================================
+    ⚠️  已验证可用的标准做法，请勿随意修改以下编码逻辑！
+    ============================================================
+    经过多轮调试（2026-02-28），以下参数组合为唯一验证成功的方案：
+
+    1. 图片格式：PNG（format="PNG"）
+       - 不可改为 JPEG —— API 会校验 Content-Type，抓包确认服务端使用 image/png
+       - 不可使用 base64 字符串 —— 会报 "expected a file, got a string"
+       - 不可使用 data URI —— 服务端不识别，返回 500
+
+    2. 图片尺寸：必须与视频分辨率完全一致（target_size）
+       - 不可缩放降采样 —— 会报 "Inpaint image must match the requested width and height"
+       - 尺寸由 _fit_image_to_target() 保证（等比缩放 + 居中裁剪）
+
+    3. 上传方式：由调用方（sora_client.py）以 multipart/form-data 文件字段上传
+       - filename="reference.png", content_type="image/png"
+       - 不可改回 application/json —— 服务端校验 input_reference 必须为 file 类型
+    ============================================================
+
+    Args:
+        image: PIL Image 对象
+        target_size: 目标分辨率字符串 "WxH"（如 "720x1280"）
+
+    Returns:
+        PNG 格式的二进制字节
+    """
+    from io import BytesIO
+
+    # 统一转换为 RGB（去除透明通道及其他模式）
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+
+    # 适配到目标分辨率（等比缩放 + 居中裁剪）
+    # ⚠️ 必须保持此尺寸不变，API 强制要求参考图片与视频分辨率完全一致
+    if target_size:
+        image = _fit_image_to_target(image, target_size)
+
+    # ⚠️ 必须使用 PNG 格式，不可改为 JPEG 或其他格式
+    buffered = BytesIO()
+    image.save(buffered, format="PNG")
+    size_kb = buffered.tell() / 1024
+    print(f"Sora: 参考图片编码为 PNG，{size_kb:.0f} KB ({image.size[0]}x{image.size[1]})")
+    return buffered.getvalue()
+
+
+class SoraVideo:
+    """
+    Sora 视频生成节点
+
+    功能：
+    - 文生视频：基于提示词生成视频
+    - 图生视频：基于参考图片和提示词生成视频
+    - 异步轮询：自动等待生成完成并下载
+    """
+
+    def __init__(self):
+        self.client = None
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        from ..models_config import SECONDS_DISPLAY_MAP, RESOLUTION_DISPLAY_MAP
+        
+        enabled_models = get_enabled_sora_models()
+        if not enabled_models:
+            enabled_models = ["请在 models_config.py 中启用至少一个 Sora 模型"]
+
+        # 构建秒数选项列表（按数字顺序排序）
+        # 格式: ["4", "8", "10", "12", "15", "25(pro)"]
+        all_seconds_display = []
+        seen_seconds = set()
+        for model_id in enabled_models:
+            supported = get_sora_supported_seconds(model_id)
+            for s in supported:
+                if s not in seen_seconds:
+                    seen_seconds.add(s)
+                    display = SECONDS_DISPLAY_MAP.get(s, str(s))
+                    all_seconds_display.append((s, display))
+        # 按秒数数值排序
+        all_seconds_display = sorted(all_seconds_display, key=lambda x: x[0])
+        seconds_options = [d for _, d in all_seconds_display] if all_seconds_display else ["4", "8", "12"]
+
+        # 构建分辨率选项列表（去重）
+        # 格式: ["720P", "1080P"]
+        seen_resolutions = set()
+        for model_id in enabled_models:
+            supported = get_sora_supported_sizes(model_id)
+            for size in supported:
+                if size in RESOLUTION_DISPLAY_MAP:
+                    res_name, _ = RESOLUTION_DISPLAY_MAP[size]
+                    seen_resolutions.add(res_name)
+        resolution_options = sorted(list(seen_resolutions)) if seen_resolutions else ["720P"]
+
+        return {
+            "required": {
+                "prompt": ("STRING", {
+                    "default": "A calico cat playing a piano on stage",
+                    "multiline": True,
+                }),
+                "模型": (enabled_models, {
+                    "default": enabled_models[0],
+                }),
+                "分辨率": (resolution_options, {
+                    "default": resolution_options[0] if resolution_options else "720P",
+                }),
+                "宽高比": (["竖屏", "横屏"], {
+                    "default": "竖屏",
+                }),
+                "视频时长": (seconds_options, {
+                    "default": seconds_options[0] if seconds_options else "4",
+                }),
+                "生成数量": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 10,
+                    "step": 1,
+                }),
+                "seed": ("INT", {
+                    "default": 0,
+                    "min": 0,
+                    "max": 0xffffffffffffffff
+                }),
+            },
+            "optional": {
+                "参考图片": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("预览视频",)
+    FUNCTION = "generate_video"
+    CATEGORY = "video/generation"
+
+    DESCRIPTION = (
+        "Sora 视频生成节点。\n"
+        "支持文生视频和图生视频，自动轮询任务状态并下载视频。\n"
+        "视频保存到 ComfyUI/output/video/ 目录。\n\n"
+        "【模型说明】\n"
+        "• sora-2：官方模型，支持 4/8/12秒、720P 分辨率\n"
+        "• sora-2-pro：增强模型，支持全时长（含25秒）、1080P 分辨率\n\n"
+        "【时长说明】\n"
+        "• 25(pro)：仅 sora-2-pro 支持的25秒时长\n\n"
+        "【分辨率说明】\n"
+        "• 720P：sora-2 和 sora-2-pro 均支持\n"
+        "• 1080P：仅 sora-2-pro 支持的高清分辨率"
+    )
+
+    def generate_video(
+        self,
+        prompt: str,
+        模型: str,
+        **kwargs,
+    ) -> Tuple[str]:
+        from ..models_config import SECONDS_DISPLAY_MAP, RESOLUTION_DISPLAY_MAP
+        
+        视频时长_display = kwargs.pop("视频时长", "4")
+        分辨率_display = kwargs.pop("分辨率", "720P")
+        宽高比 = kwargs.pop("宽高比", "竖屏")
+        生成数量 = kwargs.pop("生成数量", 1)
+        seed = kwargs.pop("seed", 0)
+        start_time = time.time()
+
+        # 解析秒数显示值（如 "25(pro)" → 25）
+        seconds = 4  # 默认
+        for actual, display in SECONDS_DISPLAY_MAP.items():
+            if display == 视频时长_display:
+                seconds = actual
+                break
+        # 如果找不到映射，尝试直接解析数字
+        if seconds == 4 and 视频时长_display != "4":
+            try:
+                seconds = int(视频时长_display.replace("(pro)", ""))
+            except ValueError:
+                seconds = 4
+
+        # 根据分辨率和宽高比确定实际分辨率值
+        分辨率 = "720x1280"  # 默认
+        for actual, (res_name, orientation) in RESOLUTION_DISPLAY_MAP.items():
+            if res_name == 分辨率_display and orientation == 宽高比:
+                分辨率 = actual
+                break
+
+        # 检查参考图片
+        ref_image = kwargs.get("参考图片")
+        ref_image_bytes = None
+        if ref_image is not None:
+            pil_images = tensor_to_pil(ref_image)
+            if pil_images:
+                ref_image_bytes = _compress_image_for_upload(pil_images[0], target_size=分辨率)
+
+        mode_str = "图生视频 (含参考图)" if ref_image_bytes else "文生视频"
+        # 获取用户友好的显示值用于日志
+        seconds_display = SECONDS_DISPLAY_MAP.get(seconds, str(seconds))
+        res_display = f"{分辨率_display} {宽高比}"
+        if 生成数量 > 1:
+            print(f"Sora: {mode_str} | 并发{生成数量}个 | {模型} | {seconds_display} | {res_display}")
+        else:
+            print(f"Sora: {mode_str} | {模型} | {seconds_display} | {res_display}")
+
+        # 校验参数兼容性
+        supported_seconds = get_sora_supported_seconds(模型)
+        if supported_seconds and seconds not in supported_seconds:
+            # 构建带标签的支持时长列表
+            supported_labels = []
+            for s in supported_seconds:
+                display = SECONDS_DISPLAY_MAP.get(s, str(s))
+                supported_labels.append(display)
+            raise ValueError(
+                f"时长 {SECONDS_DISPLAY_MAP.get(seconds, str(seconds))} 与模型 \"{模型}\" 不兼容！\n"
+                f"该模型支持的时长: {', '.join(supported_labels)}"
+            )
+        
+        supported_sizes = get_sora_supported_sizes(模型)
+        if supported_sizes and 分辨率 not in supported_sizes:
+            # 检查该分辨率是否为Pro独占
+            pro_only_sizes = ["1024x1792", "1792x1024"]
+            _, orientation = RESOLUTION_DISPLAY_MAP.get(分辨率, (分辨率, ""))
+            extra_hint = f"\n提示：1080P {orientation} 为 sora-2-pro 独占，请切换模型或选择720P。" if 分辨率 in pro_only_sizes else ""
+            raise ValueError(
+                f"分辨率 \"{分辨率_display} {宽高比}\" 与模型 \"{模型}\" 不兼容！"
+                f"支持的分辨率: {', '.join(supported_sizes)}" + extra_hint
+            )
+
+        # 准备保存路径
+        video_dir = _get_video_output_dir()
+        counter = _get_next_counter(video_dir, "sora")
+
+        # ProgressBar
+        pbar = None
+        if PROGRESS_BAR_AVAILABLE:
+            pbar = ProgressBar(生成数量 if 生成数量 > 1 else 100)
+
+        try:
+            if self.client is None:
+                self.client = SoraClient()
+
+            if 生成数量 == 1:
+                # ── 单个视频：保留详细进度（提交→轮询→下载）
+                save_path = os.path.join(video_dir, f"sora_{counter:05d}.mp4")
+                last_progress = [0]
+
+                def progress_callback(progress_pct: int):
+                    print(
+                        f"\rSora: 生成中... 进度: {progress_pct}%",
+                        end="", flush=True
+                    )
+                    if pbar is not None and progress_pct > last_progress[0]:
+                        pbar.update(progress_pct - last_progress[0])
+                        last_progress[0] = progress_pct
+
+                def on_stage(stage: str):
+                    if stage == "submitting":
+                        print("Sora: 正在提交视频生成任务...")
+                    elif stage.startswith("submitted:"):
+                        vid = stage.split(":", 1)[1]
+                        print(f"Sora: 视频任务已提交，ID: {vid}")
+                    elif stage == "polling":
+                        print("Sora: 等待视频生成...")
+                    elif stage == "downloading":
+                        print("")  # 换行（结束 \r 行）
+                        print("Sora: 视频生成完成，正在下载...")
+
+                result_path = self.client.generate_video_sync(
+                    prompt=prompt,
+                    model=模型,
+                    seconds=seconds,
+                    size=分辨率,
+                    save_path=save_path,
+                    input_reference_bytes=ref_image_bytes,
+                    seed=seed,
+                    progress_callback=progress_callback,
+                    on_stage=on_stage,
+                )
+                result_paths = [result_path]
+
+            else:
+                # ── 批量并发：同时提交多个任务
+                save_paths = [
+                    os.path.join(video_dir, f"sora_{counter + i:05d}.mp4")
+                    for i in range(生成数量)
+                ]
+                success_count = [0]
+                fail_count = [0]
+
+                def batch_progress_callback(current: int, total: int, success: bool, error_msg):
+                    if success:
+                        success_count[0] += 1
+                        print(f"Sora: 第 {current}/{total} 个视频完成 ✓")
+                    else:
+                        fail_count[0] += 1
+                        print(f"Sora: 第 {current}/{total} 个视频失败 ✗")
+                        if error_msg:
+                            print(f"原始错误详情:\n{error_msg}")
+                    if pbar is not None:
+                        pbar.update(1)
+
+                print(f"Sora: 正在并发提交 {生成数量} 个视频任务，请耐心等待...")
+                result_paths = self.client.generate_batch_videos_sync(
+                    prompt=prompt,
+                    model=模型,
+                    seconds=seconds,
+                    size=分辨率,
+                    save_paths=save_paths,
+                    input_reference_bytes=ref_image_bytes,
+                    seed=seed,
+                    progress_callback=batch_progress_callback,
+                )
+
+            elapsed = time.time() - start_time
+            time_str = f"{elapsed:.2f}s" if elapsed >= 1 else f"{elapsed:.3f}s"
+            print(f"Sora: 完成！总耗时 {time_str} | 已生成 {len(result_paths)} 个视频")
+            for p in result_paths:
+                print(f"  → {p}")
+
+            output_path = "\n".join(result_paths)
+            return (output_path,)
+
+        except ValueError as e:
+            error_msg = str(e)
+            print(f"\nSora: ❌ {error_msg}")
+            raise ValueError(error_msg) from None
+
+        except RuntimeError as e:
+            error_msg = str(e)
+            print(f"\nSora: ❌ {error_msg}")
+            raise RuntimeError(error_msg) from None
+
+        except Exception as e:
+            error_msg = str(e)
+            print(f"\nSora: ❌ {error_msg}")
+            raise type(e)(error_msg) from None
+
+        finally:
+            if self.client is not None:
+                try:
+                    balance_data = self.client.query_balance_sync()
+                    balance_info = self.client.format_balance_info(balance_data)
+                    print(f"Sora: {balance_info}")
+                except Exception:
+                    pass
@@ -0,0 +1,28 @@
+"""
+流式文本预览节点
+接收文本输入，支持 markdown 渲染，通过 ComfyUI 事件系统实时推送内容
+"""
+
+
+class StreamPreview:
+    """
+    流式 Markdown 预览节点
+    接收任意文本，在节点面板中实时渲染为 Markdown 格式
+    """
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "文本": ("STRING", {"forceInput": True}),
+            }
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("文本",)
+    FUNCTION = "preview"
+    CATEGORY = "text/preview"
+    OUTPUT_NODE = True
+
+    def preview(self, 文本: str):
+        return {"ui": {"text": [文本]}, "result": (文本,)}
@@ -0,0 +1,513 @@
+"""
+全能LLM对话助手节点
+ComfyUI 自定义节点，通过 OpenAI 兼容协议调用市面上主流的 AI 对话大模型
+支持多模态（图片输入），单轮对话，非流式输出
+
+API 密钥和地址通过插件统一配置（环境变量或 .config 文件），与 Google Gemini 节点一致
+"""
+
+import os
+import time
+import base64
+import json
+from io import BytesIO
+from typing import Optional, Tuple, List
+
+import torch
+from PIL import Image
+
+from ..utils.image_utils import tensor_to_pil
+from ..utils.config import get_api_key_or_raise, get_api_base_url, NETWORK_ROUTE_OPTIONS, get_base_url_by_route
+from ..utils.file_types import FileList
+
+# ============================================================================
+# 模型配置
+# ============================================================================
+
+SUPPORTED_MODELS = [
+    "gpt-5.5",
+    "gemini-3.1-pro-preview",
+    "deepseek-v4-pro",
+    "claude-opus-4-7",
+    "claude-opus-4-6",
+    "gemini-3.5-flash",
+    "doubao-seed-2.0-pro",
+]
+
+# 图片缩放最大尺寸
+MAX_IMAGE_DIMENSION = 1568
+
+# 图片最大文件大小（20MB）
+MAX_IMAGE_SIZE = 20 * 1024 * 1024
+
+
+class UniversalLLMChat:
+    """
+    全能LLM对话助手
+
+    功能：
+    - 通过 OpenAI 兼容协议调用主流大模型
+    - 支持多模态（图片输入）
+    - 单轮对话，非流式输出
+    - API 密钥和地址继承插件统一配置
+    """
+
+    def __init__(self):
+        self._api_key = None
+        self._base_url = None
+
+    def _ensure_config(self):
+        """延迟加载配置，首次调用时初始化"""
+        if self._api_key is None:
+            self._api_key = get_api_key_or_raise("O1KEY_API_KEY")
+            self._base_url = get_api_base_url()
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "网络线路": (NETWORK_ROUTE_OPTIONS, {
+                    "default": "全球加速"
+                }),
+                "模型": (SUPPORTED_MODELS, {
+                    "default": SUPPORTED_MODELS[0]
+                }),
+                "提示词": ("STRING", {
+                    "default": "",
+                    "multiline": True,
+                }),
+            },
+            "optional": {
+                "图片": ("IMAGE",),
+                "视频": ("VIDEO",),
+                "文件": ("FILE_LIST",),
+                "令牌": ("STRING", {
+                    "default": "",
+                    "multiline": False,
+                    "placeholder": "留空则使用默认 API Key",
+                }),
+            },
+            "hidden": {
+                "node_id": "UNIQUE_ID",
+            },
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("回复",)
+    FUNCTION = "generate"
+    CATEGORY = "text/generation"
+    OUTPUT_NODE = True
+
+    def _resize_image(self, img: Image.Image) -> Image.Image:
+        """如果图片过长边超过限制，等比缩放"""
+        w, h = img.size
+        max_dim = max(w, h)
+        if max_dim > MAX_IMAGE_DIMENSION:
+            scale = MAX_IMAGE_DIMENSION / max_dim
+            new_w, new_h = int(w * scale), int(h * scale)
+            print(f"全能LLM: 图片缩放 {w}x{h} -> {new_w}x{new_h}")
+            return img.resize((new_w, new_h), Image.Resampling.LANCZOS)
+        return img
+
+    def _image_to_data_url(self, img: Image.Image) -> str:
+        """将 PIL Image 转为 data URL（JPEG base64）"""
+        img = self._resize_image(img)
+        if img.mode in ('RGBA', 'P'):
+            img = img.convert('RGB')
+
+        for quality in [92, 82, 72, 60, 45]:
+            buf = BytesIO()
+            img.save(buf, format='JPEG', quality=quality, optimize=True)
+            data = buf.getvalue()
+            if len(data) <= MAX_IMAGE_SIZE:
+                b64 = base64.b64encode(data).decode('utf-8')
+                return f"data:image/jpeg;base64,{b64}"
+
+        b64 = base64.b64encode(data).decode('utf-8')
+        return f"data:image/jpeg;base64,{b64}"
+
+    # 文件大小限制
+    MAX_FILE_SIZE = 50 * 1024 * 1024       # 单文件 50MB
+    MAX_TOTAL_FILE_SIZE = 50 * 1024 * 1024  # 所有文件总计 50MB
+
+    # 常见 MIME 类型映射
+    MIME_MAP = {
+        ".pdf": "application/pdf",
+        ".txt": "text/plain",
+        ".md": "text/markdown",
+        ".csv": "text/csv",
+        ".json": "application/json",
+        ".py": "text/x-python",
+        ".js": "text/javascript",
+        ".html": "text/html",
+        ".xml": "application/xml",
+        ".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+        ".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+        ".pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+        ".zip": "application/zip",
+    }
+
+    # 纯文本类型，直接读取内容
+    TEXT_EXTS = {".txt", ".md", ".csv", ".json", ".py", ".js", ".ts", ".html",
+                 ".xml", ".yaml", ".yml", ".toml", ".ini", ".cfg", ".log",
+                 ".sh", ".bat", ".sql", ".css", ".scss", ".jsx", ".tsx"}
+
+    def _load_files(self, file_paths_str: str) -> List[dict]:
+        """读取文件列表，返回 content part 数组"""
+        if not file_paths_str or not file_paths_str.strip():
+            return []
+
+        paths = [p.strip() for p in file_paths_str.split(",") if p.strip()]
+        parts = []
+        total_size = 0
+
+        for path in paths:
+            if not os.path.isfile(path):
+                raise ValueError(f"文件不存在: {path}")
+
+            file_size = os.path.getsize(path)
+            if file_size > self.MAX_FILE_SIZE:
+                raise ValueError(f"文件 {os.path.basename(path)} 大小 {file_size / 1024 / 1024:.1f}MB 超过单文件 50MB 限制")
+
+            total_size += file_size
+            if total_size > self.MAX_TOTAL_FILE_SIZE:
+                raise ValueError(f"所有文件总大小超过 50MB 限制")
+
+            ext = os.path.splitext(path)[1].lower()
+            mime = self.MIME_MAP.get(ext, "application/octet-stream")
+            filename = os.path.basename(path)
+
+            if ext in self.TEXT_EXTS:
+                # 文本文件直接读取内容
+                with open(path, "r", encoding="utf-8", errors="replace") as f:
+                    text_content = f.read()
+                parts.append({
+                    "type": "text",
+                    "text": f"[文件: {filename}]\n```\n{text_content}\n```",
+                })
+            else:
+                # 二进制文件转 base64，使用 file 格式（OpenAI 兼容协议）
+                with open(path, "rb") as f:
+                    file_data = base64.b64encode(f.read()).decode("utf-8")
+                parts.append({
+                    "type": "file",
+                    "file": {
+                        "filename": filename,
+                        "file_data": f"data:{mime};base64,{file_data}",
+                    },
+                })
+
+            print(f"全能LLM: 加载文件 {filename} ({file_size / 1024:.1f}KB, {mime})")
+
+        return parts
+
+    def _build_input(
+        self,
+        prompt: str,
+        images: Optional[torch.Tensor] = None,
+        file_paths: str = "",
+        file_list: Optional[FileList] = None,
+        video=None,
+    ) -> list:
+        """构建 chat/completions 格式的 messages 数组"""
+        image_data_urls = []
+        pil_images_cache = []  # 保留 PIL Image 用于总体积重新编码
+        
+        if images is not None:
+            pil_images = tensor_to_pil(images)
+            for img in pil_images:
+                img_resized = self._resize_image(img)
+                if img_resized.mode in ('RGBA', 'P'):
+                    img_resized = img_resized.convert('RGB')
+                pil_images_cache.append(img_resized)
+                image_data_urls.append(self._image_to_data_url(img_resized))
+
+        # 多图总体积控制
+        if pil_images_cache and len(pil_images_cache) > 1:
+            total_bytes = sum(
+                len(base64.b64decode(url.split(',', 1)[1])) for url in image_data_urls
+            )
+            if total_bytes > MAX_IMAGE_SIZE:
+                print(f"全能LLM: 图片总体积 {total_bytes / 1024 / 1024:.2f}MB 超过 {MAX_IMAGE_SIZE // 1024 // 1024}MB 限制，正在压缩...")
+                
+                # 降质量
+                compressed = False
+                for quality in [80, 70, 60, 50, 40, 30, 20]:
+                    new_urls = []
+                    for img in pil_images_cache:
+                        buf = BytesIO()
+                        img.save(buf, format='JPEG', quality=quality, optimize=True)
+                        b64 = base64.b64encode(buf.getvalue()).decode('utf-8')
+                        new_urls.append(f"data:image/jpeg;base64,{b64}")
+                    total_bytes = sum(len(base64.b64decode(u.split(',', 1)[1])) for u in new_urls)
+                    if total_bytes <= MAX_IMAGE_SIZE:
+                        image_data_urls = new_urls
+                        print(f"全能LLM: 图片压缩完成，总体积 {total_bytes / 1024 / 1024:.2f}MB ({len(pil_images_cache)}张图片，质量{quality})")
+                        compressed = True
+                        break
+                
+                # 降分辨率
+                if not compressed:
+                    for scale in [0.75, 0.5, 0.35]:
+                        new_urls = []
+                        for img in pil_images_cache:
+                            w, h = img.size
+                            resized = img.resize((int(w * scale), int(h * scale)), Image.Resampling.LANCZOS)
+                            buf = BytesIO()
+                            resized.save(buf, format='JPEG', quality=20, optimize=True)
+                            b64 = base64.b64encode(buf.getvalue()).decode('utf-8')
+                            new_urls.append(f"data:image/jpeg;base64,{b64}")
+                        total_bytes = sum(len(base64.b64decode(u.split(',', 1)[1])) for u in new_urls)
+                        if total_bytes <= MAX_IMAGE_SIZE:
+                            image_data_urls = new_urls
+                            print(f"全能LLM: 图片压缩完成，总体积 {total_bytes / 1024 / 1024:.2f}MB ({len(pil_images_cache)}张图片，缩放{int(scale*100)}%)")
+                            compressed = True
+                            break
+                
+                if not compressed:
+                    print(f"全能LLM: 无法将 {len(pil_images_cache)} 张图片压缩到 {MAX_IMAGE_SIZE // 1024 // 1024}MB 以内，请减少图片数量或降低分辨率")
+                    raise ValueError(f"图片总体积 {total_bytes / 1024 / 1024:.2f}MB 超过限制，无法压缩到 {MAX_IMAGE_SIZE // 1024 // 1024}MB 以内")
+
+        # 处理视频输入（ComfyUI VIDEO 类型）
+        video_url_str = ""
+        if video is not None:
+            # 从 VIDEO 对象中提取文件路径
+            vp = None
+            if isinstance(video, dict):
+                vp = video.get("video") or video.get("path") or video.get("file") or video.get("filename")
+                if not vp:
+                    for val in video.values():
+                        if isinstance(val, str) and os.path.exists(val):
+                            vp = val
+                            break
+            elif isinstance(video, str):
+                vp = video
+            else:
+                for attr in ("video", "path", "filename"):
+                    if hasattr(video, attr):
+                        vp = getattr(video, attr)
+                        break
+                if not vp and hasattr(video, "__dict__"):
+                    for attr_val in video.__dict__.values():
+                        if isinstance(attr_val, str) and os.path.isfile(attr_val):
+                            vp = attr_val
+                            break
+
+            if not vp or not os.path.isfile(vp):
+                raise ValueError(f"视频文件不存在或路径无效: {vp}")
+
+            mime_map = {
+                ".mp4": "video/mp4", ".mpeg": "video/mpeg", ".mpg": "video/mpg",
+                ".mov": "video/quicktime", ".avi": "video/x-msvideo",
+                ".flv": "video/x-flv", ".webm": "video/webm",
+                ".wmv": "video/x-ms-wmv", ".mkv": "video/x-matroska",
+            }
+            ext = os.path.splitext(vp)[1].lower()
+            mime = mime_map.get(ext, "video/mp4")
+            file_size = os.path.getsize(vp)
+            print(f"全能LLM: 加载视频 {os.path.basename(vp)} ({file_size / 1024 / 1024:.1f}MB, {mime})")
+            with open(vp, "rb") as f:
+                b64 = base64.b64encode(f.read()).decode("utf-8")
+            video_url_str = f"data:{mime};base64,{b64}"
+
+        # 加载文件：优先使用 FILE_LIST，其次使用字符串路径
+        file_parts = []
+        if file_list:
+            for fd in file_list:
+                print(f"全能LLM: 使用文件 {fd.filename}{fd.extension} ({fd.size / 1024:.1f}KB)")
+                file_parts.append({
+                    "type": "file",
+                    "file": {
+                        "filename": fd.filename + fd.extension,
+                        "file_data": f"data:{fd.mime_type};base64,{fd.data}",
+                    },
+                })
+        elif file_paths:
+            file_parts = self._load_files(file_paths)
+
+        # 纯文本，无图片无文件无视频
+        if not image_data_urls and not file_parts and not video_url_str:
+            return [{"role": "user", "content": prompt}]
+
+        content_parts = []
+
+        # 图片
+        for url in image_data_urls:
+            content_parts.append({
+                "type": "image_url",
+                "image_url": {"url": url},
+            })
+
+        # 视频：用 image_url 类型传 data URL（Gemini OpenAI 兼容层支持此格式）
+        # 同时保留 video_url 类型作为备用（其他支持 video_url 的模型）
+        if video_url_str:
+            content_parts.append({
+                "type": "image_url",
+                "image_url": {"url": video_url_str},
+            })
+
+        # 文件
+        for fp in file_parts:
+            content_parts.append(fp)
+
+        content_parts.append({
+            "type": "text",
+            "text": prompt,
+        })
+
+        return [{"role": "user", "content": content_parts}]
+
+    @staticmethod
+    def _send_stream_token(node_id, token, done=False):
+        """通过 PromptServer 向前端推送流式 token"""
+        try:
+            from server import PromptServer
+            PromptServer.instance.send_sync(
+                "o1key.stream_token",
+                {"node_id": str(node_id), "token": token, "done": done},
+            )
+        except Exception:
+            pass
+
+    def generate(
+        self,
+        模型: str,
+        提示词: str,
+        网络线路: str = "全球加速",
+        图片: Optional[torch.Tensor] = None,
+        视频=None,
+        文件: Optional[FileList] = None,
+        令牌: str = "",
+        node_id: str = "",
+    ) -> Tuple[str]:
+        start_time = time.time()
+
+        try:
+            self._ensure_config()
+            self._base_url = get_base_url_by_route(网络线路)
+
+            # 如果用户传入了自定义令牌，则覆盖默认 API Key
+            effective_api_key = 令牌.strip() if 令牌 and 令牌.strip() else self._api_key
+
+            # 构建 input
+            input_data = self._build_input(提示词, 图片, "", 文件, 视频)
+
+            img_count = len(tensor_to_pil(图片)) if 图片 is not None else 0
+            file_count = len(文件) if 文件 else 0
+            input_desc = "文本"
+            if img_count: input_desc += f" + {img_count}张图片"
+            if 视频 is not None: input_desc += " + 视频"
+            if file_count: input_desc += f" + {file_count}个文件"
+
+            print(f"全能LLM: 模型 = {模型}")
+            print(f"全能LLM: 输入 = {input_desc}")
+
+            # 构建请求体（chat/completions 格式）
+            request_body = {
+                "model": 模型,
+                "messages": input_data,
+                "stream": True,
+            }
+
+            # 打印请求体，base64 截断显示
+            def _truncate_for_log(obj):
+                if isinstance(obj, dict):
+                    return {k: _truncate_for_log(v) for k, v in obj.items()}
+                if isinstance(obj, list):
+                    return [_truncate_for_log(i) for i in obj]
+                if isinstance(obj, str) and (obj.startswith("data:image") or obj.startswith("data:application") or obj.startswith("data:text")):
+                    return obj[:60] + f"...[{len(obj)}chars]"
+                return obj
+            print(f"全能LLM: 请求原始内容 = {json.dumps(_truncate_for_log(request_body), ensure_ascii=False)}")
+
+            # 发送请求（在独立线程中运行异步请求，避免与 ComfyUI 事件循环冲突）
+            import aiohttp
+            import asyncio
+            from concurrent.futures import ThreadPoolExecutor
+
+            async def _do_request():
+                headers = {
+                    "Content-Type": "application/json",
+                    "Authorization": f"Bearer {effective_api_key}",
+                }
+                url = f"{self._base_url}/v1/chat/completions"
+                timeout = aiohttp.ClientTimeout(total=120)
+
+                async with aiohttp.ClientSession(timeout=timeout) as session:
+                    async with session.post(url, headers=headers, json=request_body) as resp:
+                        status = resp.status
+
+                        if status != 200:
+                            body = await resp.text()
+                            try:
+                                err_data = json.loads(body)
+                                err_msg = err_data.get("error", {}).get("message", body[:200])
+                            except Exception:
+                                err_msg = body[:200]
+
+                            if status == 401:
+                                raise ValueError(f"认证失败：API Key 无效或已过期")
+                            elif status == 403:
+                                raise ValueError(f"无权访问模型 {模型}")
+                            elif status == 429:
+                                raise ValueError(f"请求频率超限，请稍后重试")
+                            elif status == 404:
+                                raise ValueError(f"模型 {模型} 不存在或 API 地址错误")
+                            else:
+                                raise RuntimeError(f"API 错误 ({status}): {err_msg}")
+
+                        # 流式读取，拼接 delta content
+                        reply_parts = []
+                        async for raw_line in resp.content:
+                            line = raw_line.decode("utf-8").strip()
+                            if not line or not line.startswith("data:"):
+                                continue
+                            data_str = line[len("data:"):].strip()
+                            if data_str == "[DONE]":
+                                break
+                            try:
+                                chunk = json.loads(data_str)
+                            except Exception:
+                                continue
+                            choices = chunk.get("choices")
+                            if not choices:
+                                continue
+                            delta = choices[0].get("delta", {})
+                            content = delta.get("content")
+                            if content:
+                                reply_parts.append(content)
+                                UniversalLLMChat._send_stream_token(node_id, content)
+
+                        UniversalLLMChat._send_stream_token(node_id, "", done=True)
+                        return "".join(reply_parts)
+
+            def _run_in_thread():
+                loop = asyncio.new_event_loop()
+                try:
+                    return loop.run_until_complete(_do_request())
+                finally:
+                    loop.close()
+
+            with ThreadPoolExecutor(max_workers=1) as pool:
+                reply = pool.submit(_run_in_thread).result()
+
+            elapsed = time.time() - start_time
+            print(f"全能LLM: 生成完成 (耗时: {elapsed:.2f}s)")
+            if reply:
+                preview = reply[:100] + "..." if len(reply) > 100 else reply
+                print(f"全能LLM: 回复预览: {preview}")
+
+            return (reply,)
+
+        except ValueError as e:
+            if str(e) == "未授权！":
+                print("全能LLM: 请联系作者授权后方可使用！")
+                raise ValueError("未授权！") from None
+            error_msg = str(e).split('\n')[0]
+            print(f"全能LLM: ❌ {error_msg}")
+            raise
+
+        except Exception as e:
+            error_msg = str(e).split('\n')[0]
+            print(f"全能LLM: ❌ {error_msg}")
+            raise RuntimeError(error_msg) from None
@@ -0,0 +1,422 @@
+"""
+Google Veo 视频生成节点
+ComfyUI 自定义节点，调用 Veo API 生成视频
+"""
+
+import os
+import re
+import time
+from typing import Optional, Tuple
+
+import torch
+
+from ..utils.image_utils import tensor_to_pil
+from ..clients.veo_client import VeoClient
+from ..models_config import (
+    get_enabled_veo_models,
+    VEO_MODELS,
+    VEO_RESOLUTION_MAP,
+)
+
+try:
+    import folder_paths
+    FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    FOLDER_PATHS_AVAILABLE = False
+
+try:
+    from comfy.utils import ProgressBar
+    PROGRESS_BAR_AVAILABLE = True
+except ImportError:
+    PROGRESS_BAR_AVAILABLE = False
+    print("⚠️ GoogleVeo: comfy.utils.ProgressBar 不可用，将只使用终端进度显示")
+
+
+def _get_video_output_dir() -> str:
+    """获取视频输出目录: ComfyUI/output/video"""
+    if FOLDER_PATHS_AVAILABLE:
+        base = folder_paths.get_output_directory()
+    else:
+        plugin_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+        base = os.path.join(os.path.dirname(os.path.dirname(plugin_dir)), "output")
+    video_dir = os.path.join(base, "video")
+    os.makedirs(video_dir, exist_ok=True)
+    return video_dir
+
+
+def _get_next_counter(directory: str, prefix: str) -> int:
+    """扫描目录，获取下一个可用的文件计数器"""
+    if not os.path.exists(directory):
+        return 1
+    pattern = re.compile(rf"^{re.escape(prefix)}_(\d+)")
+    max_counter = 0
+    for f in os.listdir(directory):
+        m = pattern.match(f)
+        if m:
+            max_counter = max(max_counter, int(m.group(1)))
+    return max_counter + 1
+
+
+def _fit_image_to_target(image, target_size: str):
+    """
+    将参考图片按 "等比缩放覆盖 + 居中裁剪" 策略适配到目标分辨率。
+    """
+    from PIL import Image as PILImage
+
+    parts = target_size.lower().split("x")
+    target_w, target_h = int(parts[0]), int(parts[1])
+
+    src_w, src_h = image.size
+    src_ratio = src_w / src_h
+    target_ratio = target_w / target_h
+
+    if abs(src_ratio - target_ratio) < 0.01 and src_w <= target_w and src_h <= target_h:
+        return image
+
+    print(f"Veo: 参考图片 {src_w}x{src_h} (比例 {src_ratio:.2f}) → 目标 {target_w}x{target_h} (比例 {target_ratio:.2f})")
+
+    resample = PILImage.Resampling.LANCZOS if hasattr(PILImage, "Resampling") else PILImage.LANCZOS
+
+    if src_ratio > target_ratio:
+        scale = target_h / src_h
+        new_w = round(src_w * scale)
+        new_h = target_h
+        image = image.resize((new_w, new_h), resample=resample)
+        left = (new_w - target_w) // 2
+        image = image.crop((left, 0, left + target_w, target_h))
+    else:
+        scale = target_w / src_w
+        new_w = target_w
+        new_h = round(src_h * scale)
+        image = image.resize((new_w, new_h), resample=resample)
+        top = (new_h - target_h) // 2
+        image = image.crop((0, top, target_w, top + target_h))
+
+    print(f"Veo: 参考图片已适配为 {image.size[0]}x{image.size[1]}")
+    return image
+
+
+def _compress_image_to_bytes(image, target_size: Optional[str] = None) -> bytes:
+    """
+    将 PIL Image 适配目标分辨率并编码为 PNG 字节
+    """
+    from io import BytesIO
+
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+
+    if target_size:
+        image = _fit_image_to_target(image, target_size)
+
+    buffered = BytesIO()
+    image.save(buffered, format="PNG")
+    size_kb = buffered.tell() / 1024
+    print(f"Veo: 参考图片编码为 PNG，{size_kb:.0f} KB ({image.size[0]}x{image.size[1]})")
+    return buffered.getvalue()
+
+
+class GoogleVeo:
+    """
+    Google Veo 视频生成节点
+
+    功能：
+    - 文生视频：基于提示词生成视频
+    - 图生视频：基于首帧/尾帧/参考图生成视频
+    - 异步轮询：自动等待生成完成并下载
+    """
+
+    def __init__(self):
+        self.client = None
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        enabled_models = get_enabled_veo_models()
+        if not enabled_models:
+            enabled_models = ["请在 models_config.py 中启用 Veo 模型"]
+
+        # 分辨率选项
+        resolution_options = ["720p", "1080p", "4K"]
+
+        # 宽高比选项
+        aspect_ratio_options = ["16:9", "9:16"]
+
+        # 视频秒数选项
+        seconds_options = ["4", "6", "8"]
+
+        return {
+            "required": {
+                "prompt": ("STRING", {
+                    "default": "A calico cat playing a piano on stage",
+                    "multiline": True,
+                }),
+                "模型": (enabled_models, {
+                    "default": enabled_models[0] if enabled_models else "Veo3.1",
+                }),
+                "分辨率": (resolution_options, {
+                    "default": "720p",
+                }),
+                "宽高比": (aspect_ratio_options, {
+                    "default": "9:16",
+                }),
+                "视频时长": (seconds_options, {
+                    "default": "8",
+                }),
+                "seed": ("INT", {
+                    "default": 0,
+                    "min": 0,
+                    "max": 0xffffffffffffffff,
+                }),
+                "生成数量": ("INT", {
+                    "default": 1,
+                    "min": 1,
+                    "max": 10,
+                    "step": 1,
+                }),
+            },
+            "optional": {
+                "首帧": ("IMAGE",),
+                "尾帧": ("IMAGE",),
+                "参考图": ("IMAGE",),
+            },
+        }
+
+    RETURN_TYPES = ("STRING",)
+    RETURN_NAMES = ("预览视频",)
+    FUNCTION = "generate_video"
+    CATEGORY = "video/generation"
+
+    DESCRIPTION = (
+        "Google Veo 视频生成节点。\n"
+        "支持文生视频和图生视频（图生视频支持首帧、尾帧、参考图）。\n"
+        "视频保存到 ComfyUI/output/video/ 目录。\n\n"
+        "【模型说明】\n"
+        "• Veo3.1：Google 最新视频生成模型\n\n"
+        "【分辨率说明】\n"
+        "• 720p：标清\n"
+        "• 1080p：高清\n"
+        "• 4K：超高清\n\n"
+        "【时长说明】\n"
+        "• 4秒：短视频\n"
+        "• 6秒：标准\n"
+        "• 8秒：长视频（默认）\n\n"
+        "【图生视频说明】\n"
+        "• 首帧：视频开始的第一帧图像\n"
+        "• 尾帧：视频结束时的最后一帧图像\n"
+        "• 参考图：参考图像（与首帧/尾帧配合使用）\n"
+        "• 至少需要提供首帧或参考图之一"
+    )
+
+    def generate_video(
+        self,
+        prompt: str,
+        模型: str,
+        **kwargs,
+    ) -> Tuple[str]:
+        分辨率 = kwargs.pop("分辨率", "720p")
+        宽高比 = kwargs.pop("宽高比", "9:16")
+        视频时长 = kwargs.pop("视频时长", "8")
+        seed = kwargs.pop("seed", 0)
+        生成数量 = kwargs.pop("生成数量", 1)
+        start_time = time.time()
+
+        # 解析视频时长
+        seconds = int(视频时长)
+
+        # 解析分辨率和宽高比，映射到模型名称
+        size_key = f"{分辨率}_{宽高比}"
+        actual_size = VEO_RESOLUTION_MAP.get(size_key)
+        if not actual_size:
+            # 默认值
+            actual_size = "720x1280"  # 720p 9:16
+
+        # 检查是否有参考图输入
+        首帧 = kwargs.get("首帧")
+        尾帧 = kwargs.get("尾帧")
+        参考图 = kwargs.get("参考图")
+
+        has_image = 首帧 is not None or 尾帧 is not None or 参考图 is not None
+
+        # 根据是否有图片选择模型前缀
+        if has_image:
+            model_prefix = "veo3.1"
+        else:
+            model_prefix = "veo3.1"
+
+        # 构建完整模型名称
+        # 格式: veo3.1-portrait / veo3.1-landscape / veo3.1-portrait-fl / veo3.1-landscape-fl 等
+        if 分辨率 == "720p":
+            res_suffix = ""
+            if 宽高比 == "9:16":
+                orientation = "portrait"
+            else:
+                orientation = "landscape"
+        elif 分辨率 == "1080p":
+            res_suffix = "-hd"
+            if 宽高比 == "9:16":
+                orientation = "portrait"
+            else:
+                orientation = "landscape"
+        else:  # 4K
+            res_suffix = "-4k"
+            if 宽高比 == "9:16":
+                orientation = "portrait"
+            else:
+                orientation = "landscape"
+
+        # 图生视频添加 -fl 后缀
+        if has_image:
+            model_suffix = f"-{orientation}-fl{res_suffix}"
+        else:
+            model_suffix = f"-{orientation}{res_suffix}"
+
+        model = f"{model_prefix}{model_suffix}"
+
+        # 准备图片字节
+        first_frame_bytes = None
+        last_frame_bytes = None
+        reference_bytes = None
+
+        if 首帧 is not None:
+            pil_images = tensor_to_pil(首帧)
+            if pil_images:
+                first_frame_bytes = _compress_image_to_bytes(pil_images[0], target_size=actual_size)
+
+        if 尾帧 is not None:
+            pil_images = tensor_to_pil(尾帧)
+            if pil_images:
+                last_frame_bytes = _compress_image_to_bytes(pil_images[0], target_size=actual_size)
+
+        if 参考图 is not None:
+            pil_images = tensor_to_pil(参考图)
+            if pil_images:
+                reference_bytes = _compress_image_to_bytes(pil_images[0], target_size=actual_size)
+
+        mode_str = "图生视频" if has_image else "文生视频"
+        print(f"Veo: {mode_str} | 并发{生成数量}个 | 模型: {model} | {seconds}秒 | {分辨率} {宽高比}")
+
+        # 准备保存路径
+        video_dir = _get_video_output_dir()
+        counter = _get_next_counter(video_dir, "veo")
+
+        # ProgressBar
+        pbar = None
+        if PROGRESS_BAR_AVAILABLE:
+            pbar = ProgressBar(生成数量 if 生成数量 > 1 else 100)
+
+        try:
+            if self.client is None:
+                self.client = VeoClient()
+
+            if 生成数量 == 1:
+                save_path = os.path.join(video_dir, f"veo_{counter:05d}.mp4")
+                last_progress = [0]
+
+                def progress_callback(progress_pct: int):
+                    print(
+                        f"\rVeo: 生成中... 进度: {progress_pct}%",
+                        end="", flush=True
+                    )
+                    if pbar is not None and progress_pct > last_progress[0]:
+                        pbar.update(progress_pct - last_progress[0])
+                        last_progress[0] = progress_pct
+
+                def on_stage(stage: str):
+                    if stage == "submitting":
+                        print("Veo: 正在提交视频生成任务...")
+                    elif stage.startswith("submitted:"):
+                        vid = stage.split(":", 1)[1]
+                        print(f"Veo: 视频任务已提交，ID: {vid}")
+                    elif stage == "polling":
+                        print("Veo: 等待视频生成...")
+                    elif stage == "downloading":
+                        print("")
+                        print("Veo: 视频生成完成，正在下载...")
+
+                result_path = self.client.generate_video_sync(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=actual_size,
+                    save_path=save_path,
+                    first_frame_bytes=first_frame_bytes,
+                    last_frame_bytes=last_frame_bytes,
+                    reference_bytes=reference_bytes,
+                    seed=seed,
+                    progress_callback=progress_callback,
+                    on_stage=on_stage,
+                )
+                result_paths = [result_path]
+
+            else:
+                save_paths = [
+                    os.path.join(video_dir, f"veo_{counter + i:05d}.mp4")
+                    for i in range(生成数量)
+                ]
+                success_count = [0]
+
+                def batch_progress_callback(current: int, total: int, success: bool, error_msg):
+                    if success:
+                        success_count[0] += 1
+                        print(f"Veo: 第 {current}/{total} 个视频完成 ✓")
+                    else:
+                        print(f"Veo: 第 {current}/{total} 个视频失败 ✗")
+                        if error_msg:
+                            print(f"原始错误详情:\n{error_msg}")
+                    if pbar is not None:
+                        pbar.update(1)
+
+                print(f"Veo: 正在并发提交 {生成数量} 个视频任务，请耐心等待...")
+                result_paths = self.client.generate_batch_videos_sync(
+                    prompt=prompt,
+                    model=model,
+                    seconds=seconds,
+                    size=actual_size,
+                    save_paths=save_paths,
+                    first_frame_bytes=first_frame_bytes,
+                    last_frame_bytes=last_frame_bytes,
+                    reference_bytes=reference_bytes,
+                    seed=seed,
+                    progress_callback=batch_progress_callback,
+                )
+
+            elapsed = time.time() - start_time
+            time_str = f"{elapsed:.2f}s" if elapsed >= 1 else f"{elapsed:.3f}s"
+            print(f"Veo: 完成！总耗时 {time_str} | 已生成 {len(result_paths)} 个视频")
+            for p in result_paths:
+                print(f"  → {p}")
+
+            output_path = "\n".join(result_paths)
+            return (output_path,)
+
+        except ValueError as e:
+            error_msg = str(e)
+            print(f"\nVeo: ❌ {error_msg}")
+            raise ValueError(error_msg) from None
+
+        except RuntimeError as e:
+            error_msg = str(e)
+            print(f"\nVeo: ❌ {error_msg}")
+            raise RuntimeError(error_msg) from None
+
+        except Exception as e:
+            error_msg = str(e)
+            print(f"\nVeo: ❌ {error_msg}")
+            raise type(e)(error_msg) from None
+
+        finally:
+            if self.client is not None:
+                try:
+                    balance_data = self.client.query_balance_sync()
+                    balance_info = self.client.format_balance_info(balance_data)
+                    print(f"Veo: {balance_info}")
+                except Exception:
+                    pass
+
+
+NODE_CLASS_MAPPINGS = {
+    "GoogleVeo": GoogleVeo,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "GoogleVeo": "Google Veo - ab",
+}
@@ -0,0 +1,90 @@
+"""
+视频预览节点
+接收 VIDEO 类型，在前端内嵌播放器预览
+"""
+
+import os
+import io
+
+try:
+    import folder_paths
+    FOLDER_PATHS_AVAILABLE = True
+except ImportError:
+    FOLDER_PATHS_AVAILABLE = False
+
+
+def _get_output_dir() -> str:
+    if FOLDER_PATHS_AVAILABLE:
+        return folder_paths.get_output_directory()
+    plugin_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+    return os.path.join(os.path.dirname(os.path.dirname(plugin_dir)), "output")
+
+
+class VideoPreview:
+
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "视频": ("VIDEO",),
+            },
+        }
+
+    RETURN_TYPES = ()
+    OUTPUT_NODE = True
+    FUNCTION = "preview"
+    CATEGORY = "comfyui_o1key/Utils"
+
+    def preview(self, 视频) -> dict:
+        # 用官方接口取文件路径
+        source = 视频.get_stream_source()
+
+        if isinstance(source, io.BytesIO):
+            # BytesIO 情况：写到 output/video/ 临时文件
+            output_dir = os.path.join(_get_output_dir(), "video")
+            os.makedirs(output_dir, exist_ok=True)
+            filename = "preview_tmp.mp4"
+            tmp_path = os.path.join(output_dir, filename)
+            source.seek(0)
+            with open(tmp_path, "wb") as f:
+                f.write(source.read())
+            subfolder = "video"
+        else:
+            video_path = source
+            output_dir = _get_output_dir()
+            abs_video = os.path.abspath(video_path)
+            abs_output = os.path.abspath(output_dir)
+
+            if abs_video.startswith(abs_output):
+                rel_path = os.path.relpath(abs_video, abs_output)
+                subfolder = os.path.dirname(rel_path).replace("\\", "/")
+                filename = os.path.basename(rel_path)
+            else:
+                # 文件在 output 目录外，复制一份
+                target_dir = os.path.join(output_dir, "video")
+                os.makedirs(target_dir, exist_ok=True)
+                filename = os.path.basename(abs_video)
+                target_path = os.path.join(target_dir, filename)
+                if not os.path.exists(target_path):
+                    import shutil
+                    shutil.copy2(abs_video, target_path)
+                subfolder = "video"
+
+        return {
+            "ui": {
+                "videos": [{
+                    "filename": filename,
+                    "subfolder": subfolder,
+                    "type": "output",
+                }],
+            }
+        }
+
+
+NODE_CLASS_MAPPINGS = {
+    "VideoPreview": VideoPreview,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "VideoPreview": "预览视频",
+}
@@ -1,3 +1,4 @@
 aiohttp>=3.9.0
 Pillow>=10.0.0
 requests>=2.31.0
+rembg[cpu]>=2.0.50
@@ -0,0 +1,73 @@
+"""Git integration checks for the sidebar updater."""
+
+import importlib.util
+import subprocess
+import tempfile
+import unittest
+from pathlib import Path
+
+
+UPDATER_PATH = Path(__file__).resolve().parents[1] / "utils" / "updater.py"
+spec = importlib.util.spec_from_file_location("o1key_updater_under_test", UPDATER_PATH)
+updater = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(updater)
+
+
+class UpdaterTests(unittest.TestCase):
+    def setUp(self):
+        self.temp = tempfile.TemporaryDirectory()
+        self.addCleanup(self.temp.cleanup)
+        root = Path(self.temp.name)
+        self.remote = root / "remote.git"
+        self.author = root / "author"
+        self.install = root / "install"
+        self.git(root, "init", "--bare", str(self.remote))
+        self.git(root, "clone", str(self.remote), str(self.author))
+        self.git(self.author, "config", "user.email", "[email protected]")
+        self.git(self.author, "config", "user.name", "Updater Test")
+        self.git(self.author, "switch", "-c", "main")
+        (self.author / "requirements.txt").write_text("requests>=2\n", encoding="utf-8")
+        (self.author / "version.txt").write_text("1\n", encoding="utf-8")
+        self.commit_and_push()
+        self.git(root, "clone", "--branch", "main", str(self.remote), str(self.install))
+        updater.PLUGIN_DIR = self.install
+
+    def git(self, cwd, *args):
+        return subprocess.run(["git", *args], cwd=cwd, check=True, capture_output=True, text=True).stdout.strip()
+
+    def commit_and_push(self):
+        self.git(self.author, "add", ".")
+        self.git(self.author, "commit", "-m", "test update")
+        self.git(self.author, "push", "origin", "main")
+
+    def test_fast_forward_and_requirements_change(self):
+        self.assertFalse(updater.update_package()["updated"])
+        (self.author / "version.txt").write_text("2\n", encoding="utf-8")
+        (self.author / "requirements.txt").write_text("requests>=3\n", encoding="utf-8")
+        self.commit_and_push()
+        result = updater.update_package()
+        self.assertTrue(result["updated"])
+        self.assertTrue(result["requirements_changed"])
+        self.assertEqual((self.install / "version.txt").read_text(encoding="utf-8"), "2\n")
+
+    def test_local_changes_are_preserved(self):
+        (self.install / "version.txt").write_text("local\n", encoding="utf-8")
+        with self.assertRaisesRegex(updater.UpdateError, "本地修改"):
+            updater.update_package()
+        self.assertEqual((self.install / "version.txt").read_text(encoding="utf-8"), "local\n")
+
+    def test_diverged_branch_is_rejected(self):
+        self.git(self.install, "config", "user.email", "[email protected]")
+        self.git(self.install, "config", "user.name", "Updater Test")
+        (self.install / "version.txt").write_text("local commit\n", encoding="utf-8")
+        self.git(self.install, "add", ".")
+        self.git(self.install, "commit", "-m", "local")
+        (self.author / "version.txt").write_text("remote commit\n", encoding="utf-8")
+        self.commit_and_push()
+        with self.assertRaisesRegex(updater.UpdateError, "已分叉"):
+            updater.update_package()
+        self.assertEqual((self.install / "version.txt").read_text(encoding="utf-8"), "local commit\n")
+
+
+if __name__ == "__main__":
+    unittest.main()
@@ -1,97 +1,73 @@
@echo off
-chcp 65001 > nul
-echo ====================================
-echo Comfyui_o1key 插件更新工具
-echo ====================================
-echo.
-
-:: 检查是否在 Git 仓库中
-if not exist ".git" (
-    echo [错误] 当前目录不是 Git 仓库
-    echo 请确保插件是通过 git clone 安装的
-    pause
-    exit /b 1
-)
-
-:: 保存当前版本
-if exist "version.txt" (
-    set /p OLD_VERSION=<version.txt
-    echo 当前版本: %OLD_VERSION%
-) else (
-    set OLD_VERSION=未知
-    echo 当前版本: 未知
-)
+setlocal EnableDelayedExpansion
+title comfyui_o1key Updater
+cd /d "%~dp0"

 echo.
-echo [1/4] 检查远程更新...
-git fetch origin
-
-:: 检查是否有更新
-git status -uno | findstr "Your branch is behind" > nul
-if %errorlevel% equ 0 (
-    echo 发现新版本！
-) else (
-    echo 已是最新版本
-    echo.
-    choice /C YN /M "是否继续检查依赖更新？"
-    if errorlevel 2 goto :end
-)
-
+echo  [ comfyui_o1key Updater ]
 echo.
-echo [2/4] 备份配置文件...
-if exist ".config" (
-    copy /Y ".config" ".config.backup" > nul
-    echo 已备份 .config 到 .config.backup
-)

+:: Check git
+where git >nul 2>&1
+if errorlevel 1 ( set "ERR=Git not found in PATH." & goto :fail )
+
+:: Check repo
+if not exist ".git" ( set "ERR=Not a git repo. Place this file in the plugin root." & goto :fail )
+
+:: Save old hash
+for /f %%i in ('git rev-parse --short HEAD 2^>nul') do set "OLD=%%i"
+
+:: Backup .config
+if exist ".config" copy /y ".config" ".config.bak" >nul 2>&1
+
+:: Fetch
+echo  Fetching...
+git fetch origin >nul 2>&1
+if errorlevel 1 ( set "ERR=Network error. Check GitHub access." & goto :fail )
+
+:: Already up to date?
+for /f %%i in ('git rev-parse HEAD 2^>nul')        do set "LOCAL=%%i"
+for /f %%i in ('git rev-parse origin/main 2^>nul') do set "REMOTE=%%i"
+if "%LOCAL%"=="%REMOTE%" ( goto :uptodate )
+
+:: Switch branch & force reset
+git branch --list main | findstr "main" >nul 2>&1
+if errorlevel 1 ( git checkout -b main origin/main >nul 2>&1 ) else ( git checkout main >nul 2>&1 )
+
+echo  Updating...
+git reset --hard origin/main >nul 2>&1
+if errorlevel 1 ( set "ERR=git reset failed." & goto :fail )
+git clean -fd -e ".config" -e ".config.bak" >nul 2>&1
+
+:: Restore .config
+if exist ".config.bak" ( copy /y ".config.bak" ".config" >nul 2>&1 & del /f /q ".config.bak" >nul 2>&1 )
+
+for /f %%i in ('git rev-parse --short HEAD 2^>nul') do set "NEW=%%i"
 echo.
-echo [3/4] 拉取最新代码...
-git pull origin main
-if %errorlevel% neq 0 (
-    echo [错误] 代码更新失败，请检查网络连接或手动解决冲突
-    pause
-    exit /b 1
-)
-
-:: 恢复配置文件
-if exist ".config.backup" (
-    copy /Y ".config.backup" ".config" > nul
-    del ".config.backup"
-    echo 已恢复配置文件
-)
-
+echo  +---------------------------+
+echo  ^|  SUCCESS                  ^|
+echo  ^|  %OLD%  ->  %NEW%         ^|
+echo  ^|  Restart ComfyUI          ^|
+echo  +---------------------------+
 echo.
-echo [4/4] 更新依赖包...
-python -m pip install -r requirements.txt --upgrade --quiet
-if %errorlevel% neq 0 (
-    echo [警告] 依赖包更新失败，请手动运行: pip install -r requirements.txt
-)
+pause & exit /b 0

+:uptodate
+if exist ".config.bak" ( copy /y ".config.bak" ".config" >nul 2>&1 & del /f /q ".config.bak" >nul 2>&1 )
 echo.
-echo ====================================
-echo 更新完成！
-echo ====================================
-
-:: 显示新版本
-if exist "version.txt" (
-    set /p NEW_VERSION=<version.txt
-    echo 新版本: %NEW_VERSION%
-)
-
-:: 显示最近更新日志
-if exist "CHANGELOG.md" (
-    echo.
-    echo 最近更新内容:
-    echo -----------------------------------
-    powershell -Command "Get-Content CHANGELOG.md -TotalCount 20"
-    echo -----------------------------------
-)
-
+echo  +---------------------------+
+echo  ^|  Already up to date       ^|
+echo  ^|  %LOCAL:~0,7%  (no change)       ^|
+echo  +---------------------------+
 echo.
-echo 请重启 ComfyUI 以使更改生效
-echo.
-pause
-goto :end
+pause & exit /b 0

-:end
-exit /b 0
+:fail
+if exist ".config.bak" ( copy /y ".config.bak" ".config" >nul 2>&1 & del /f /q ".config.bak" >nul 2>&1 )
+echo.
+echo  +---------------------------+
+echo  ^|  FAILED                   ^|
+echo  ^|  %ERR%
+echo  +---------------------------+
+echo.
+pause & exit /b 1
@@ -1,101 +0,0 @@
-#!/bin/bash
-
-# 设置颜色输出
-GREEN='\033[0;32m'
-YELLOW='\033[1;33m'
-RED='\033[0;31m'
-NC='\033[0m' # No Color
-
-echo "===================================="
-echo "Comfyui_o1key 插件更新工具"
-echo "===================================="
-echo ""
-
-# 检查是否在 Git 仓库中
-if [ ! -d ".git" ]; then
-    echo -e "${RED}[错误] 当前目录不是 Git 仓库${NC}"
-    echo "请确保插件是通过 git clone 安装的"
-    exit 1
-fi
-
-# 保存当前版本
-if [ -f "version.txt" ]; then
-    OLD_VERSION=$(cat version.txt)
-    echo "当前版本: $OLD_VERSION"
-else
-    OLD_VERSION="未知"
-    echo "当前版本: 未知"
-fi
-
-echo ""
-echo "[1/4] 检查远程更新..."
-git fetch origin
-
-# 检查是否有更新
-LOCAL=$(git rev-parse @)
-REMOTE=$(git rev-parse @{u})
-
-if [ $LOCAL != $REMOTE ]; then
-    echo -e "${GREEN}发现新版本！${NC}"
-else
-    echo -e "${GREEN}已是最新版本${NC}"
-    read -p "是否继续检查依赖更新？(y/n) " -n 1 -r
-    echo
-    if [[ ! $REPLY =~ ^[Yy]$ ]]; then
-        exit 0
-    fi
-fi
-
-echo ""
-echo "[2/4] 备份配置文件..."
-if [ -f ".config" ]; then
-    cp .config .config.backup
-    echo "已备份 .config 到 .config.backup"
-fi
-
-echo ""
-echo "[3/4] 拉取最新代码..."
-if git pull origin main; then
-    echo -e "${GREEN}代码更新成功${NC}"
-else
-    echo -e "${RED}[错误] 代码更新失败，请检查网络连接或手动解决冲突${NC}"
-    exit 1
-fi
-
-# 恢复配置文件
-if [ -f ".config.backup" ]; then
-    mv .config.backup .config
-    echo "已恢复配置文件"
-fi
-
-echo ""
-echo "[4/4] 更新依赖包..."
-if python3 -m pip install -r requirements.txt --upgrade --quiet; then
-    echo -e "${GREEN}依赖包更新成功${NC}"
-else
-    echo -e "${YELLOW}[警告] 依赖包更新失败，请手动运行: pip install -r requirements.txt${NC}"
-fi
-
-echo ""
-echo "===================================="
-echo -e "${GREEN}更新完成！${NC}"
-echo "===================================="
-
-# 显示新版本
-if [ -f "version.txt" ]; then
-    NEW_VERSION=$(cat version.txt)
-    echo "新版本: $NEW_VERSION"
-fi
-
-# 显示最近更新日志
-if [ -f "CHANGELOG.md" ]; then
-    echo ""
-    echo "最近更新内容:"
-    echo "-----------------------------------"
-    head -n 20 CHANGELOG.md
-    echo "-----------------------------------"
-fi
-
-echo ""
-echo "请重启 ComfyUI 以使更改生效"
-echo ""
@@ -15,8 +15,7 @@ from .file_utils import (
    load_images_from_folder,
    pair_images_indexed,
    pair_images_cartesian,
-    generate_output_filename,
-    generate_batch_output_filenames,
+    generate_timestamp_filename,
    save_image,
    get_folder_image_count
 )
@@ -32,8 +31,7 @@ __all__ = [
    'load_images_from_folder',
    'pair_images_indexed',
    'pair_images_cartesian',
-    'generate_output_filename',
-    'generate_batch_output_filenames',
+    'generate_timestamp_filename',
    'save_image',
    'get_folder_image_count'
 ]
@@ -0,0 +1,174 @@
+"""
+颜色去背景工具模块
+基于颜色距离计算实现精确可控的背景移除，不依赖 AI 模型。
+
+支持模式：
+  - white: 白色背景去除
+  - white-preserve: 白色背景但保护浅色前景物体
+  - corner: 自动采样四角颜色作为背景色
+  - color: 指定任意颜色去除
+"""
+
+import numpy as np
+from PIL import Image
+
+
+def background_to_alpha(
+    image: Image.Image,
+    bg_color: tuple = (255, 255, 255),
+    tolerance: float = 8.0,
+    feather: float = 45.0,
+    strength: float = 1.0,
+    min_alpha: int = 2,
+) -> Image.Image:
+    """
+    将纯色背景转为透明。
+
+    对白色背景使用 white-to-alpha 恢复算法，保持彩色文字和抗锯齿边缘清晰。
+    对其他颜色使用欧氏距离计算。
+    """
+    rgba = np.asarray(image.convert("RGBA")).astype(np.float32)
+    rgb = rgba[:, :, :3] / 255.0
+    existing_alpha = rgba[:, :, 3] / 255.0
+    bg = np.array(bg_color, dtype=np.float32) / 255.0
+
+    if max(bg_color) >= 245 and min(bg_color) >= 245:
+        alpha = (1.0 - np.min(rgb, axis=2)) * float(strength)
+        if tolerance > 0:
+            dist = np.linalg.norm((1.0 - rgb) * 255.0, axis=2)
+            gate = np.clip(
+                (dist - float(tolerance)) / max(1.0, float(feather) * 0.25),
+                0.0, 1.0,
+            )
+            alpha *= gate
+    else:
+        dist = np.linalg.norm((rgb - bg) * 255.0, axis=2)
+        denom = max(1.0, float(feather))
+        alpha = np.clip((dist - float(tolerance)) / denom, 0.0, 1.0)
+        alpha *= float(strength)
+
+    alpha = np.clip(alpha, 0.0, 1.0) * existing_alpha
+    alpha[alpha < (float(min_alpha) / 255.0)] = 0.0
+
+    # 从 alpha 混合中恢复前景色，避免白边
+    out_rgb = rgb.copy()
+    mask = alpha > 1e-6
+    out_rgb[mask] = (rgb[mask] - bg * (1.0 - alpha[mask, None])) / alpha[mask, None]
+    out_rgb = np.clip(out_rgb, 0.0, 1.0)
+
+    out = np.dstack([
+        (out_rgb * 255.0).astype(np.uint8),
+        (alpha * 255.0).astype(np.uint8),
+    ])
+    return Image.fromarray(out, "RGBA")
+
+
+def corner_color(image: Image.Image, sample: int = 12) -> tuple:
+    """采样图片四角像素的中位数颜色，用于自动检测背景色。"""
+    rgb = np.asarray(image.convert("RGB"))
+    h, w = rgb.shape[:2]
+    sample = max(1, min(sample, h, w))
+    patches = [
+        rgb[:sample, :sample],
+        rgb[:sample, w - sample:],
+        rgb[h - sample:, :sample],
+        rgb[h - sample:, w - sample:],
+    ]
+    merged = np.concatenate([p.reshape(-1, 3) for p in patches], axis=0)
+    return tuple(np.median(merged, axis=0).astype(int))
+
+
+# PLACEHOLDER_PRESERVE
+
+def preserve_light_foreground_to_alpha(
+    image: Image.Image,
+    tolerance: float = 10.0,
+    preserve_opacity: float = 0.72,
+    min_area_ratio: float = 0.00025,
+) -> Image.Image:
+    """
+    白底去除 + 浅色前景保护。
+
+    适用于前景包含白色/浅色物体（白盘子、白帆、白色包装）的场景。
+    使用 OpenCV 连通区域分析保护大面积浅色前景结构。
+    如果 OpenCV 不可用，回退到普通 white-to-alpha。
+    """
+    base = background_to_alpha(image, (255, 255, 255), tolerance=tolerance)
+    try:
+        import cv2
+    except ImportError:
+        return base
+
+    rgb_u8 = np.asarray(image.convert("RGB"))
+    h, w = rgb_u8.shape[:2]
+    dist = np.sqrt(np.sum((255.0 - rgb_u8.astype(np.float32)) ** 2, axis=2))
+    rough = (dist > float(tolerance)).astype(np.uint8) * 255
+
+    kernel_open = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (5, 5))
+    kernel_close = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (17, 17))
+    rough = cv2.morphologyEx(rough, cv2.MORPH_OPEN, kernel_open, iterations=1)
+    rough = cv2.morphologyEx(rough, cv2.MORPH_CLOSE, kernel_close, iterations=2)
+
+    count, labels, stats, _ = cv2.connectedComponentsWithStats(rough, 8)
+    keep = np.zeros_like(rough)
+    min_area = max(24, int(w * h * float(min_area_ratio)))
+    for idx in range(1, count):
+        if stats[idx, cv2.CC_STAT_AREA] >= min_area:
+            keep[labels == idx] = 255
+
+    # PLACEHOLDER_FLOOD
+
+    flood = keep.copy()
+    ff_mask = np.zeros((h + 2, w + 2), dtype=np.uint8)
+    cv2.floodFill(flood, ff_mask, (0, 0), 255)
+    filled = cv2.bitwise_or(keep, cv2.bitwise_not(flood))
+    soft = cv2.GaussianBlur(filled, (0, 0), 5).astype(np.float32) / 255.0
+
+    near_kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (29, 29))
+    near = cv2.dilate(
+        (dist > (float(tolerance) * 0.65)).astype(np.uint8) * 255,
+        near_kernel, iterations=1,
+    )
+    near = cv2.GaussianBlur(near, (0, 0), 8).astype(np.float32) / 255.0
+    lift = np.minimum(soft, near) * float(preserve_opacity)
+
+    arr = np.asarray(base.convert("RGBA")).copy()
+    alpha = arr[:, :, 3].astype(np.float32) / 255.0
+    alpha = np.maximum(alpha, lift)
+    alpha[alpha < (2.0 / 255.0)] = 0.0
+
+    original = np.asarray(image.convert("RGB"))
+    very_light = (np.mean(original, axis=2) > 224) & (lift > 0.12)
+    arr[:, :, :3][very_light] = original[very_light]
+    arr[:, :, 3] = np.clip(alpha * 255.0, 0, 255).astype(np.uint8)
+    return Image.fromarray(arr, "RGBA")
+
+
+def remove_background(
+    image: Image.Image,
+    mode: str = "white",
+    bg_color: tuple = (255, 255, 255),
+    tolerance: float = 8.0,
+    feather: float = 45.0,
+    strength: float = 1.0,
+) -> Image.Image:
+    """
+    统一入口：根据模式移除背景。
+
+    mode:
+      - white: 白色背景去除
+      - white-preserve: 白底 + 保护浅色前景
+      - corner: 自动采样四角颜色
+      - color: 使用指定 bg_color
+    """
+    if mode == "white":
+        return background_to_alpha(image, (255, 255, 255), tolerance, feather, strength)
+    elif mode == "white-preserve":
+        return preserve_light_foreground_to_alpha(image, tolerance)
+    elif mode == "corner":
+        bg = corner_color(image)
+        return background_to_alpha(image, bg, tolerance, feather, strength)
+    elif mode == "color":
+        return background_to_alpha(image, bg_color, tolerance, feather, strength)
+    else:
+        return image.convert("RGBA")
@@ -12,6 +12,24 @@ PLUGIN_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 CONFIG_FILE = os.path.join(PLUGIN_ROOT, ".config")


+# ============ API 基础配置 ============
+# 所有 API 客户端的统一基础 URL
+# 可通过环境变量 O1KEY_API_BASE_URL 覆盖
+DEFAULT_API_BASE_URL = "https://api.o1key.com"
+
+# 异步 API 基础 URL（用于异步提交+轮询模式）
+# 可通过环境变量 O1KEY_ASYNC_API_BASE_URL 覆盖
+DEFAULT_ASYNC_API_BASE_URL = "https://cf-api.o1key.com"
+
+# ============ 网络线路配置 ============
+NETWORK_ROUTES = {
+    "全球加速": "https://api.o1key.cn",
+    "CF加速": "https://cf-api.o1key.com",
+    "美国直连": "https://api.o1key.com",
+}
+NETWORK_ROUTE_OPTIONS = ["全球加速", "CF加速", "美国直连"]
+
+
 def load_config(config_path: Optional[str] = None) -> Dict[str, str]:
    """
    从配置文件加载所有配置项
@@ -61,36 +79,16 @@ def load_config(config_path: Optional[str] = None) -> Dict[str, str]:
 def get_api_key(key_name: str = "O1KEY_API_KEY") -> Optional[str]:
    """
    获取 API 密钥
-    优先级：环境变量（推荐） > .config 文件（向后兼容）
-    
+    从 .config 文件读取
+
    Args:
        key_name: 密钥名称，默认为 O1KEY_API_KEY
-    
+
    Returns:
        API 密钥字符串，如果未找到则返回 None
-    
-    Raises:
-        ValueError: 如果未找到 API 密钥
-    
-    Example:
-        >>> api_key = get_api_key()
-        >>> if api_key is None:
-        ...     raise ValueError("API key not found")
    """
-    # 1. 优先从环境变量读取（推荐方式）
-    api_key = os.environ.get(key_name)
-    
-    if api_key:
-        return api_key
-    
-    # 2. 从 .config 文件读取（向后兼容，已弃用）
    config = load_config()
-    api_key = config.get(key_name)
-    
-    if api_key:
-        return api_key
-    
-    return None
+    return config.get(key_name)


 def get_api_key_or_raise(key_name: str = "O1KEY_API_KEY") -> str:
@@ -112,3 +110,44 @@ def get_api_key_or_raise(key_name: str = "O1KEY_API_KEY") -> str:
        raise ValueError("未授权！")
    
    return api_key
+
+
+def get_api_base_url() -> str:
+    """
+    获取 API 基础 URL
+    从 .config 文件读取，如果未配置则使用默认值
+
+    Returns:
+        API 基础 URL 字符串
+    """
+    config = load_config()
+    base_url = config.get("O1KEY_API_BASE_URL")
+
+    if base_url:
+        return base_url.rstrip('/')
+
+    return DEFAULT_API_BASE_URL
+
+
+def get_async_api_base_url() -> str:
+    """
+    获取异步 API 基础 URL
+    从 .config 文件读取，如果未配置则使用默认值
+
+    Returns:
+        异步 API 基础 URL 字符串
+    """
+    config = load_config()
+    base_url = config.get("O1KEY_ASYNC_API_BASE_URL")
+
+    if base_url:
+        return base_url.rstrip('/')
+
+    return DEFAULT_ASYNC_API_BASE_URL
+
+
+def get_base_url_by_route(route: str) -> str:
+    """根据网络线路选项返回对应域名，未匹配则走 config 垫底"""
+    if isinstance(route, (list, tuple)):
+        route = route[0] if route else None
+    return NETWORK_ROUTES.get(route, get_api_base_url())
@@ -0,0 +1,82 @@
+"""
+文件数据类型定义
+用于在 ComfyUI 节点间传递文件数据
+"""
+
+from typing import NamedTuple, List
+
+
+class FileData(NamedTuple):
+    """
+    单个文件数据，用于节点间传递
+
+    Attributes:
+        path: 文件完整路径
+        filename: 文件名（不含扩展名）
+        extension: 文件扩展名（如 .pdf）
+        mime_type: MIME 类型
+        data: Base64 编码的文件内容
+        size: 文件大小（字节）
+    """
+    path: str
+    filename: str
+    extension: str
+    mime_type: str
+    data: str
+    size: int
+
+
+# FILE_LIST 类型：FileData 的列表，用于多文件传递
+# ComfyUI 自定义类型名，节点 RETURN_TYPES / INPUT_TYPES 中使用 "FILE_LIST"
+FileList = List[FileData]
+
+
+# 支持的文件 MIME 类型映射（与 universal_llm.py 的 MIME_MAP 保持一致）
+DOCUMENT_MIME_TYPES = {
+    ".pdf": "application/pdf",
+    ".txt": "text/plain",
+    ".md": "text/markdown",
+    ".csv": "text/csv",
+    ".json": "application/json",
+    ".py": "text/x-python",
+    ".js": "text/javascript",
+    ".ts": "text/javascript",
+    ".html": "text/html",
+    ".xml": "application/xml",
+    ".yaml": "text/plain",
+    ".yml": "text/plain",
+    ".toml": "text/plain",
+    ".ini": "text/plain",
+    ".cfg": "text/plain",
+    ".log": "text/plain",
+    ".sh": "text/plain",
+    ".bat": "text/plain",
+    ".sql": "text/plain",
+    ".css": "text/plain",
+    ".scss": "text/plain",
+    ".jsx": "text/javascript",
+    ".tsx": "text/javascript",
+    ".docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
+    ".xlsx": "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+    ".pptx": "application/vnd.openxmlformats-officedocument.presentationml.presentation",
+    ".zip": "application/zip",
+    ".png": "image/png",
+    ".jpg": "image/jpeg",
+    ".jpeg": "image/jpeg",
+    ".webp": "image/webp",
+    ".wav": "audio/wav",
+    ".mp3": "audio/mpeg",
+    ".mp4": "video/mp4",
+}
+
+# 单文件大小上限：50MB
+FILE_SIZE_LIMIT = 50 * 1024 * 1024
+
+# 所有文件总大小上限：50MB
+TOTAL_FILE_SIZE_LIMIT = 50 * 1024 * 1024
+
+# 兼容旧代码
+FILE_SIZE_LIMITS = {
+    ".pdf": FILE_SIZE_LIMIT,
+    ".txt": FILE_SIZE_LIMIT,
+}
@@ -4,8 +4,11 @@
 """

 import os
+import re
 import uuid
 import time
+import random
+from datetime import datetime
 from itertools import product
 from pathlib import Path
 from typing import List, Tuple, Optional, NamedTuple
@@ -13,6 +16,45 @@ from typing import List, Tuple, Optional, NamedTuple
 from PIL import Image


+def _get_server_port() -> Optional[int]:
+    """获取当前 ComfyUI 实例的端口号，失败返回 None"""
+    try:
+        import comfy.cli_args
+        port = getattr(comfy.cli_args.args, 'port', None) or getattr(comfy.cli_args, 'server_port', None) or getattr(comfy.cli_args, 'port', None)
+        if port is not None:
+            return int(port)
+    except Exception:
+        pass
+    # 备用：从 listen 环境变量或命令行参数尝试
+    try:
+        import sys
+        for arg in sys.argv:
+            if '--port' in arg or '--listen-port' in arg:
+                parts = arg.split('=')
+                if len(parts) == 2:
+                    return int(parts[1].strip())
+                elif arg in ('--port', '--listen-port'):
+                    idx = sys.argv.index(arg)
+                    if idx + 1 < len(sys.argv):
+                        return int(sys.argv[idx + 1])
+    except Exception:
+        pass
+    return None
+
+
+def _get_port_suffix() -> str:
+    """
+    返回非默认端口的后缀字符串（如 "_8189"），默认端口 8188 或获取失败时返回空字符串。
+    """
+    try:
+        port = _get_server_port()
+        if port is not None and port != 8188:
+            return f"_{port}"
+    except Exception:
+        pass
+    return ""
+
+
 # 支持的图片格式
 SUPPORTED_IMAGE_EXTENSIONS = {'.jpg', '.jpeg', '.png', '.webp', '.bmp', '.gif'}

@@ -133,6 +175,77 @@ def pair_images_indexed(
    return list(zip(*non_empty_lists))


+def pair_images_by_name(
+    *image_lists: List[ImageInfo]
+) -> List[Tuple[ImageInfo, ...]]:
+    """
+    按文件名配对（同名匹配）
+
+    取所有文件夹中文件名（不含扩展名）的交集，按文件名字母升序排列后配对。
+    只有在所有文件夹中都存在同名文件，该文件名才会被纳入配对。
+    扩展名不同的文件（如 1.jpg 与 1.png）视为同名。
+
+    Args:
+        *image_lists: 多个 ImageInfo 列表
+
+    Returns:
+        配对后的元组列表，按文件名字母升序排列
+
+    Raises:
+        ValueError: 所有文件夹之间没有任何相同文件名时抛出
+
+    Example:
+        >>> list_a = [ImageInfo(filename="1", ...), ImageInfo(filename="2", ...)]
+        >>> list_b = [ImageInfo(filename="1", ...), ImageInfo(filename="3", ...)]
+        >>> pairs = pair_images_by_name(list_a, list_b)
+        >>> # [(list_a[0], list_b[0])]  # 只有 "1" 匹配
+    """
+    if not image_lists:
+        return []
+
+    non_empty_lists = [lst for lst in image_lists if lst]
+    if not non_empty_lists:
+        return []
+
+    # 单文件夹直接返回（无需配对）
+    if len(non_empty_lists) == 1:
+        return [(img,) for img in non_empty_lists[0]]
+
+    # 为每个文件夹建立 filename（stem）-> ImageInfo 的映射
+    name_maps = [
+        {img.filename: img for img in lst}
+        for lst in non_empty_lists
+    ]
+
+    # 取所有文件夹文件名的交集
+    common_names = set(name_maps[0].keys())
+    for nm in name_maps[1:]:
+        common_names &= set(nm.keys())
+
+    if not common_names:
+        # 收集各文件夹的文件名示例，帮助用户排查问题
+        folder_samples = []
+        for i, nm in enumerate(name_maps):
+            sample = sorted(nm.keys())[:3]
+            sample_str = "、".join(f'"{n}"' for n in sample)
+            folder_samples.append(f"文件夹{i + 1}：{sample_str}")
+        samples_info = "\n".join(folder_samples)
+        raise ValueError(
+            f"所有文件夹中没有找到任何同名图片，无法进行配对！\n"
+            f"请确保各文件夹内存在文件名相同的图片后重试。\n"
+            f"（文件名比较不含扩展名，例如「1.jpg」与「1.png」视为同名）\n\n"
+            f"各文件夹当前文件名示例：\n{samples_info}"
+        )
+
+    # 按文件名字母升序排列，保证顺序稳定
+    sorted_names = sorted(common_names, key=lambda x: x.lower())
+
+    return [
+        tuple(nm[name] for nm in name_maps)
+        for name in sorted_names
+    ]
+
+
 def pair_images_cartesian(
    *image_lists: List[ImageInfo]
 ) -> List[Tuple[ImageInfo, ...]]:
@@ -166,100 +279,36 @@ def pair_images_cartesian(
    return list(product(*non_empty_lists))


-def generate_output_filename(
-    source_images: List[ImageInfo],
-    batch_index: int,
-    output_folder: str,
-    extension: str = ".png",
-    task_id: Optional[str] = None
-) -> str:
+def generate_timestamp_filename(output_folder: str, prefix: str = "", extension: str = ".png", port_suffix: str = "") -> str:
    """
-    生成智能输出文件名
-    
-    基于源图片文件名生成输出文件名，使用任务ID和时间戳确保并发安全。
-    
-    Args:
-        source_images: 源图片信息列表
-        batch_index: 批次索引（从 0 开始）
-        output_folder: 输出文件夹路径
-        extension: 输出文件扩展名
-        task_id: 任务唯一标识符（用于并发场景）
-    
-    Returns:
-        完整的输出文件路径
-    
-    Example:
-        >>> # 单图片: hello.png -> hello_task0_12345_000.png
-        >>> # 多图片: hello.png + ref.png -> hello_ref_task0_12345_000.png
-        >>> # 并发安全：每个任务有唯一的 task_id 和时间戳
-    """
-    # 构建基础文件名
-    if len(source_images) == 1:
-        base_name = source_images[0].filename
-    else:
-        # 多个源图片，组合文件名
-        names = [info.filename for info in source_images]
-        base_name = "_".join(names)
-    
-    # 确保输出文件夹存在
-    output_path = Path(output_folder)
-    output_path.mkdir(parents=True, exist_ok=True)
-    
-    # 生成唯一性标识
-    if task_id is None:
-        # 如果没有提供 task_id，使用 UUID 前8位
-        task_id = str(uuid.uuid4())[:8]
-    
-    # 使用时间戳（毫秒级）增加唯一性
-    timestamp = int(time.time() * 1000) % 100000  # 精确到毫秒的后5位
-    
-    # 生成文件名：基础名_任务ID_时间戳_批次索引
-    filename = f"{base_name}_{task_id}_{timestamp:05d}_{batch_index:03d}{extension}"
-    full_path = output_path / filename
-    
-    # 极小概率的冲突处理
-    counter = 1
-    while full_path.exists():
-        filename = f"{base_name}_{task_id}_{timestamp:05d}_{batch_index:03d}_{counter}{extension}"
-        full_path = output_path / filename
-        counter += 1
-    
-    return str(full_path)
+    生成基于时间戳的文件名，确保按文件名排序 = 按生成时间排序。

+    格式：{prefix}{HHMMSS_YYYYMMDD_mmm}{port_suffix}{extension}
+    例如：161700_20260322_001.png 或 去除ai_161700_20260322_001.png

-def generate_batch_output_filenames(
-    source_images: List[ImageInfo],
-    count: int,
-    output_folder: str,
-    extension: str = ".png",
-    task_id: Optional[str] = None
-) -> List[str]:
-    """
-    批量生成输出文件名
-    
    Args:
-        source_images: 源图片信息列表
-        count: 需要生成的文件名数量
-        output_folder: 输出文件夹路径
-        extension: 输出文件扩展名
-        task_id: 任务唯一标识符（用于并发场景）
-    
+        output_folder: 输出目录
+        prefix: 文件名前缀（如 "去除ai_"）
+        extension: 文件扩展名（如 ".png"）
+        port_suffix: 端口后缀（如 "_8189"），为空时自动获取
+
    Returns:
-        输出文件路径列表
+        完整文件路径
    """
-    filenames = []
-    
-    for i in range(count):
-        filename = generate_output_filename(
-            source_images=source_images,
-            batch_index=i,
-            output_folder=output_folder,
-            extension=extension,
-            task_id=task_id
-        )
-        filenames.append(filename)
-    
-    return filenames
+    Path(output_folder).mkdir(parents=True, exist_ok=True)
+    if not port_suffix:
+        port_suffix = _get_port_suffix()
+
+    date_part = datetime.now().strftime("%Y%m%d")
+    time_part = datetime.now().strftime("%H%M%S")
+    ms = random.randint(0, 999)
+
+    while True:
+        filename = f"{prefix}{time_part}_{date_part}_{ms:03d}{port_suffix}{extension}"
+        full_path = Path(output_folder) / filename
+        if not full_path.exists():
+            return str(full_path)
+        ms = (ms + 1) % 1000


 def save_image(
@@ -290,8 +339,6 @@ def save_image(
        if image.mode != 'RGB':
            image = image.convert('RGB')
        image.save(output_path, quality=quality)
-    elif ext == '.png':
-        image.save(output_path)
    elif ext == '.webp':
        image.save(output_path, quality=quality)
    else:
@@ -0,0 +1,195 @@
+"""
+统一 HTTP 错误处理 & 退避重试模块
+
+使用方式：
+  1. 对于 aiohttp 请求，用 async_request_with_retry() 包裹 POST/GET 调用
+  2. 对于已拿到 status code 的场景，调用 raise_for_status() 抛出友好错误
+
+新增生图/视频节点时，请统一使用本模块处理 HTTP 错误。
+"""
+
+import asyncio
+import json
+import random
+from typing import Any, Optional
+
+import aiohttp
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# 状态码 → 用户友好文案
+# ═══════════════════════════════════════════════════════════════════════════════
+
+HTTP_ERROR_MESSAGES = {
+    429: "模型速率超限或额度不足！",
+    502: "网关超时。请重试或将网络切换为美国直连",
+    503: "模型超载。请稍后重试！",
+    504: "网关超时。请稍后重试。",
+}
+
+# 错误内容关键词 → 用户友好文案（优先于状态码匹配）
+ERROR_CONTENT_MESSAGES = {
+    "Your request was rejected by the safety system": "请求被安全系统拦截：请调整提示词，避免敏感、违规、血腥、色情、仇恨、未成年人或真实人物等高风险内容。",
+    "safety system": "请求被安全系统拦截：请调整提示词，避免敏感、违规、血腥、色情、仇恨、未成年人或真实人物等高风险内容。",
+    "unexpected end of JSON input": "通常重试能解决；反复出现就降低分辨率、数量或换网络线路。",
+    "The current model has a high load": "模型过载，请稍后重试！",
+    "system error": "系统错误，请稍后重试。",
+}
+
+# 可退避重试的状态码
+RETRYABLE_STATUS_CODES = {429, 502, 503, 504, 524}
+
+# 退避重试默认参数
+DEFAULT_MAX_RETRIES = 3
+DEFAULT_BASE_DELAY = 2.0      # 首次重试等待秒数
+DEFAULT_MAX_DELAY = 30.0      # 最大等待秒数
+DEFAULT_BACKOFF_FACTOR = 2.0  # 指数退避因子
+
+
+def _extract_message_from_payload(payload: Any) -> str:
+    if isinstance(payload, str):
+        text = payload.strip()
+        if not text:
+            return ""
+        if text.startswith("{") or text.startswith("["):
+            try:
+                return _extract_message_from_payload(json.loads(text))
+            except Exception:
+                return text
+        return text
+
+    if not isinstance(payload, dict):
+        return ""
+
+    error = payload.get("error")
+    if isinstance(error, dict):
+        for key in ("message", "msg", "detail", "reason"):
+            value = error.get(key)
+            if value:
+                return _extract_message_from_payload(value)
+    elif error:
+        return _extract_message_from_payload(error)
+
+    for key in ("message", "msg", "detail", "reason", "error_message"):
+        value = payload.get(key)
+        if value:
+            return _extract_message_from_payload(value)
+
+    for key in ("data", "result", "response", "output"):
+        value = payload.get(key)
+        nested = _extract_message_from_payload(value)
+        if nested:
+            return nested
+
+    return ""
+
+
+def extract_structured_error_message(raw_message: str) -> str:
+    if not isinstance(raw_message, str):
+        return ""
+    text = raw_message.strip()
+    if not (text.startswith("{") or text.startswith("[")):
+        return ""
+    return _extract_message_from_payload(text)
+
+
+def get_friendly_message(status_code: int, raw_message: str = "") -> str:
+    """根据状态码/错误内容返回友好文案，未匹配则返回原始信息"""
+    if status_code == 524:
+        return "Gateway timed out while waiting for upstream image generation. Please retry, lower resolution/count, or switch network route."
+    if raw_message:
+        structured_message = extract_structured_error_message(raw_message)
+        message_for_matching = structured_message or raw_message
+        raw_message_lower = message_for_matching.lower()
+        for keyword, friendly_msg in ERROR_CONTENT_MESSAGES.items():
+            if keyword.lower() in raw_message_lower:
+                return friendly_msg
+        if structured_message:
+            return structured_message
+    if status_code == 500:
+        return "服务器返回 500：上游生成失败或服务端临时异常。请稍后重试；如果多次出现，请降低分辨率/数量，或调整提示词。"
+    friendly = HTTP_ERROR_MESSAGES.get(status_code)
+    if friendly:
+        return friendly
+    return raw_message or f"请求失败 ({status_code})"
+
+
+def raise_for_status(status_code: int, raw_message: str = "", prefix: str = ""):
+    """根据状态码抛出带友好文案的 RuntimeError"""
+    friendly = get_friendly_message(status_code, raw_message)
+    full_msg = f"{prefix}{friendly}" if prefix else friendly
+    raise RuntimeError(full_msg)
+
+
+def is_retryable(status_code: int) -> bool:
+    return status_code in RETRYABLE_STATUS_CODES
+
+
+def _compute_delay(attempt: int, base_delay: float, max_delay: float, backoff_factor: float) -> float:
+    """计算第 attempt 次重试的等待时间（含 jitter）"""
+    delay = base_delay * (backoff_factor ** attempt)
+    delay = min(delay, max_delay)
+    jitter = random.uniform(0, delay * 0.3)
+    return delay + jitter
+
+
+async def async_request_with_retry(
+    session: aiohttp.ClientSession,
+    method: str,
+    url: str,
+    *,
+    max_retries: int = DEFAULT_MAX_RETRIES,
+    base_delay: float = DEFAULT_BASE_DELAY,
+    max_delay: float = DEFAULT_MAX_DELAY,
+    backoff_factor: float = DEFAULT_BACKOFF_FACTOR,
+    prefix: str = "",
+    **request_kwargs,
+) -> aiohttp.ClientResponse:
+    """
+    带退避重试的 aiohttp 请求。
+
+    仅对 RETRYABLE_STATUS_CODES (429/502/503/504/524) 进行重试。
+    超过最大重试次数后抛出友好 RuntimeError。
+    成功时返回 response 对象（调用者需在 async with 外自行处理 body）。
+
+    用法示例：
+        resp = await async_request_with_retry(session, "POST", url, json=body, headers=headers)
+        data = await resp.json()
+    """
+    last_status: Optional[int] = None
+    last_message = ""
+
+    for attempt in range(max_retries + 1):
+        try:
+            resp = await session.request(method, url, **request_kwargs)
+        except (aiohttp.ClientError, asyncio.TimeoutError) as e:
+            if attempt < max_retries:
+                delay = _compute_delay(attempt, base_delay, max_delay, backoff_factor)
+                print(f"{prefix}网络错误，{delay:.1f}s 后重试 ({attempt+1}/{max_retries})...")
+                await asyncio.sleep(delay)
+                continue
+            raise RuntimeError(f"{prefix}网络错误: {e}") from None
+
+        if resp.status == 200:
+            return resp
+
+        last_status = resp.status
+        try:
+            last_message = await resp.text()
+        except Exception:
+            last_message = ""
+
+        if is_retryable(resp.status) and attempt < max_retries:
+            delay = _compute_delay(attempt, base_delay, max_delay, backoff_factor)
+            friendly = get_friendly_message(resp.status)
+            print(f"{prefix}{friendly} {delay:.1f}s 后重试 ({attempt+1}/{max_retries})...")
+            await asyncio.sleep(delay)
+            continue
+
+        break
+
+    if last_status and last_status in HTTP_ERROR_MESSAGES:
+        raise_for_status(last_status, raw_message=last_message, prefix=prefix)
+
+    friendly = get_friendly_message(last_status or 0, last_message)
+    raise RuntimeError(f"{prefix}{friendly}")
@@ -5,7 +5,8 @@

 import base64
 from io import BytesIO
-from typing import List
+import json
+from typing import Callable, List, Tuple

 import numpy as np
 import torch
@@ -49,36 +50,36 @@ def tensor_to_pil(tensor: torch.Tensor) -> List[Image.Image]:
 def pil_to_tensor(images: List[Image.Image]) -> torch.Tensor:
    """
    将 PIL Image 列表转换为 ComfyUI 的 Tensor
-    
+
    Args:
        images: PIL Image 列表
-    
+
    Returns:
        形状为 [B, H, W, C] 的张量，值范围 [0, 1]
-    
+
    Example:
        >>> pil_images = [Image.open("test.png")]
        >>> tensor = pil_to_tensor(pil_images)
        >>> print(tensor.shape)  # [1, H, W, 3]
    """
    tensors = []
-    
+
    for img in images:
        # 确保是 RGB 模式
        if img.mode != 'RGB':
            img = img.convert('RGB')
-        
+
        # 转换为 numpy 数组
        img_array = np.array(img).astype(np.float32)
-        
+
        # 转换值范围从 [0, 255] 到 [0, 1]
        img_array = img_array / 255.0
-        
+
        tensors.append(img_array)
-    
+
    # 堆叠为批次
    batch_tensor = np.stack(tensors, axis=0)
-    
+
    # 转换为 torch tensor
    return torch.from_numpy(batch_tensor)

@@ -110,6 +111,180 @@ def encode_image_to_base64(image: Image.Image, format: str = "PNG") -> str:
    return base64.b64encode(img_bytes).decode('utf-8')


+_MAX_REQUEST_BODY_BYTES = 50 * 1024 * 1024  # 50MB 请求体上限
+
+
+def _encode_image_to_base64_with_quality(image: Image.Image, quality: int) -> str:
+    buffered = BytesIO()
+    working = image
+    if working.mode != 'RGB':
+        working = working.convert('RGB')
+
+    working.save(
+        buffered,
+        format="JPEG",
+        quality=quality,
+        optimize=True,
+        subsampling=2,
+    )
+    return base64.b64encode(buffered.getvalue()).decode('utf-8')
+
+
+def encode_images_for_request_body_limit(
+    images: List[Image.Image],
+    build_body: Callable[[List[Tuple[str, str]]], dict],
+    max_body_bytes: int = _MAX_REQUEST_BODY_BYTES,
+) -> List[Tuple[str, str]]:
+    """
+    为请求体编码图片，并保证完整 JSON 请求体不超过 max_body_bytes。
+
+    策略：
+    - 先按原始 PNG 编码估算完整请求体；
+    - 若超过限制，改用 JPEG 质量压缩，逐步降低 quality；
+    - 全程不缩放图片尺寸。
+
+    Returns:
+        [(mime_type, base64), ...]
+    """
+    encoded = [("image/png", encode_image_to_base64(img, format="PNG")) for img in images]
+    body_size = len(json.dumps(build_body(encoded)).encode("utf-8"))
+    if body_size <= max_body_bytes:
+        return encoded
+
+    for quality in [95, 90, 85, 80, 75, 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 5, 1]:
+        encoded = [
+            ("image/jpeg", _encode_image_to_base64_with_quality(img, quality))
+            for img in images
+        ]
+        body_size = len(json.dumps(build_body(encoded)).encode("utf-8"))
+        if body_size <= max_body_bytes:
+            print(
+                f"输入图片已通过 JPEG 质量压缩控制请求体积: "
+                f"quality={quality}, 请求体积={body_size / 1024 / 1024:.2f}MB "
+                f"(限制 {max_body_bytes / 1024 / 1024:.0f}MB)"
+            )
+            return encoded
+
+    raise ValueError(
+        f"请求体超过 {max_body_bytes / 1024 / 1024:.0f}MB，"
+        "即使压缩到最低图片质量仍无法满足限制；请减少参考图数量或输入图片内容复杂度"
+    )
+
+
+_MAX_IMAGE_BYTES = 10 * 1024 * 1024  # 10MB 单张图片上限
+
+
+def _encode_image_to_bytes(image: Image.Image, format: str = "PNG", quality: int = None) -> bytes:
+    buffered = BytesIO()
+    working = image
+
+    if format.upper() == "JPEG" and working.mode != 'RGB':
+        working = working.convert('RGB')
+    elif working.mode == 'RGBA':
+        working = working.convert('RGB')
+
+    save_kwargs = {"format": format}
+    if quality is not None:
+        save_kwargs.update({
+            "quality": quality,
+            "optimize": True,
+            "subsampling": 2,
+        })
+
+    working.save(buffered, **save_kwargs)
+    return buffered.getvalue()
+
+
+def encode_images_for_image_size_limit(
+    images: List[Image.Image],
+    max_image_bytes: int = _MAX_IMAGE_BYTES,
+) -> List[Tuple[str, str]]:
+    """
+    将图片编码为 base64，并保证每张编码前的图片文件体积不超过 max_image_bytes。
+
+    策略：
+    - 先尝试 PNG 原图尺寸编码；
+    - 单张超过限制时，改用 JPEG 质量压缩；
+    - 全程不缩放图片尺寸。
+
+    Returns:
+        [(mime_type, base64), ...]
+    """
+    encoded = []
+
+    for idx, img in enumerate(images, start=1):
+        png_bytes = _encode_image_to_bytes(img, format="PNG")
+        if len(png_bytes) <= max_image_bytes:
+            encoded.append(("image/png", base64.b64encode(png_bytes).decode('utf-8')))
+            continue
+
+        for quality in [95, 90, 85, 80, 75, 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 5, 1]:
+            jpg_bytes = _encode_image_to_bytes(img, format="JPEG", quality=quality)
+            if len(jpg_bytes) <= max_image_bytes:
+                print(
+                    f"输入图片 {idx} 已通过 JPEG 质量压缩控制单图体积: "
+                    f"quality={quality}, 图片体积={len(jpg_bytes) / 1024 / 1024:.2f}MB "
+                    f"(限制 {max_image_bytes / 1024 / 1024:.0f}MB)，尺寸保持 {img.width}x{img.height}"
+                )
+                encoded.append(("image/jpeg", base64.b64encode(jpg_bytes).decode('utf-8')))
+                break
+        else:
+            raise ValueError(
+                f"输入图片 {idx} 超过 {max_image_bytes / 1024 / 1024:.0f}MB，"
+                "即使压缩到最低图片质量仍无法满足限制；请减少图片内容复杂度或手动处理图片"
+            )
+
+    return encoded
+
+
+def encode_image_to_base64_limited(
+    image: Image.Image,
+    format: str = "PNG",
+    max_bytes: int = _MAX_IMAGE_BYTES,
+) -> str:
+    """
+    将 PIL Image 编码为 base64，若超过 max_bytes 则自动缩放直到满足限制。
+
+    策略：等比缩放，每轮缩小到上一轮的 80%，最多 10 轮。
+
+    Args:
+        image: PIL Image 对象
+        format: 图像格式，默认 PNG
+        max_bytes: base64 字符串最大字节数，默认 10MB
+
+    Returns:
+        base64 编码的字符串（保证 <= max_bytes）
+    """
+    working = image
+    if working.mode == 'RGBA':
+        working = working.convert('RGB')
+
+    for attempt in range(10):
+        buffered = BytesIO()
+        working.save(buffered, format=format)
+        b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
+
+        if len(b64) <= max_bytes:
+            if attempt > 0:
+                print(
+                    f"图片已自动缩放: {image.width}x{image.height} → "
+                    f"{working.width}x{working.height} "
+                    f"({len(b64) / 1024 / 1024:.2f}MB)"
+                )
+            return b64
+
+        # 缩放到 80%
+        scale = 0.8
+        new_w = max(1, int(working.width * scale))
+        new_h = max(1, int(working.height * scale))
+        working = working.resize((new_w, new_h), Image.Resampling.LANCZOS)
+
+    # 兜底：返回最后一次编码结果
+    buffered = BytesIO()
+    working.save(buffered, format=format)
+    return base64.b64encode(buffered.getvalue()).decode('utf-8')
+
+
 def decode_base64_to_pil(base64_string: str) -> Image.Image:
    """
    将 base64 字符串解码为 PIL Image
@@ -190,4 +365,4 @@ def parse_batch_prompts(prompt: str) -> List[str]:
    if not filtered_prompts:
        raise ValueError("批量提示词模式下，所有提示词都为空，请至少提供一个有效的提示词")
    
-    return filtered_prompts
+    return filtered_prompts
@@ -0,0 +1,675 @@
+import asyncio
+import base64
+import json
+import time
+from io import BytesIO
+from typing import Any, Callable, Dict, Iterable, List, Optional, Sequence, Tuple
+
+import aiohttp
+from PIL import Image
+
+from .http_error import (
+    DEFAULT_BACKOFF_FACTOR,
+    DEFAULT_BASE_DELAY,
+    DEFAULT_MAX_DELAY,
+    DEFAULT_MAX_RETRIES,
+    RETRYABLE_STATUS_CODES,
+    _compute_delay,
+    extract_structured_error_message,
+    get_friendly_message,
+)
+from ..clients.gemini_client import GeminiAPIClient
+
+
+_MAX_BODY_BYTES = 20_000_000
+_BODY_TARGET_BYTES = int(_MAX_BODY_BYTES * 0.8)
+_SUBMIT_ENDPOINT = "/async/v1/generateImage"
+_TASK_ENDPOINT = "/async/v1/tasks/{task_id}"
+_POLL_SCHEDULE = [5.0, 20.0]
+_POLL_INTERVAL = 3.0
+_MAX_WAIT_SECONDS = 900.0
+_INTERRUPT_STEP = 0.2
+_RUNNING_PROGRESS_MAX = 0.99
+_POLL_LOG_ENABLED = False
+
+_SUCCESS_STATUSES = {"success", "succeed", "succeeded", "completed", "done", "finished"}
+_FAILURE_STATUSES = {
+    "failure",
+    "fail",
+    "failed",
+    "error",
+    "expired",
+    "timeout",
+    "timed_out",
+    "cancel",
+    "canceled",
+    "cancelled",
+    "rejected",
+}
+_RUNNING_STATUSES = {
+    "submitted",
+    "queued",
+    "pending",
+    "running",
+    "processing",
+    "in_progress",
+    "in-progress",
+    "created",
+}
+
+
+def _headers(api_key: str) -> Dict[str, str]:
+    return {
+        "Authorization": f"Bearer {api_key}",
+        "Content-Type": "application/json",
+    }
+
+
+def _json_dumps(body: Dict[str, Any]) -> str:
+    return json.dumps(body, ensure_ascii=False, separators=(",", ":"))
+
+
+def _json_size(body: Dict[str, Any]) -> int:
+    return len(_json_dumps(body).encode("utf-8"))
+
+
+def _scale_images(images: List[Image.Image], scale: float) -> List[Image.Image]:
+    if scale >= 1.0:
+        return images
+    scaled = []
+    for img in images:
+        new_w = max(1, int(img.width * scale))
+        new_h = max(1, int(img.height * scale))
+        scaled.append(img.resize((new_w, new_h), Image.Resampling.LANCZOS))
+    return scaled
+
+
+def _encode_image_data_url(
+    image: Image.Image,
+    image_format: str,
+    quality: Optional[int] = None,
+) -> str:
+    buffered = BytesIO()
+    working = image
+    fmt = image_format.upper()
+    save_kwargs = {"format": fmt}
+
+    if fmt == "JPEG":
+        if working.mode != "RGB":
+            working = working.convert("RGB")
+        save_kwargs.update({"quality": quality or 90, "optimize": True, "subsampling": 2})
+        mime_type = "image/jpeg"
+    else:
+        if working.mode == "RGBA":
+            working = working.convert("RGB")
+        mime_type = "image/png"
+
+    working.save(buffered, **save_kwargs)
+    encoded = base64.b64encode(buffered.getvalue()).decode("ascii")
+    return f"data:{mime_type};base64,{encoded}"
+
+
+def _encode_image_data_urls(
+    images: Sequence[Image.Image],
+    image_format: str,
+    quality: Optional[int] = None,
+) -> List[str]:
+    return [_encode_image_data_url(img, image_format, quality) for img in images]
+
+
+def _fit_image_data_urls_to_body_limit(
+    images: Sequence[Image.Image],
+    build_body: Callable[[List[str]], Dict[str, Any]],
+) -> Tuple[List[str], List[Image.Image], str, int]:
+    working_images = list(images)
+    image_urls = _encode_image_data_urls(working_images, "PNG")
+    body_size = _json_size(build_body(image_urls))
+    if body_size <= _BODY_TARGET_BYTES:
+        return image_urls, working_images, "PNG", body_size
+
+    for _ in range(10):
+        if body_size <= _BODY_TARGET_BYTES:
+            break
+        ratio = _BODY_TARGET_BYTES / max(body_size, 1)
+        scale = min(0.98, ratio ** 0.5)
+        working_images = _scale_images(working_images, scale)
+        image_urls = _encode_image_data_urls(working_images, "PNG")
+        body_size = _json_size(build_body(image_urls))
+
+    return image_urls, working_images, "PNG", body_size
+
+
+def _shorten_base64_for_log(value: Any, max_len: int = 160) -> Any:
+    if isinstance(value, dict):
+        result = {}
+        for key, item in value.items():
+            if key in ("data", "b64_json", "base64", "image_base64") and isinstance(item, str) and len(item) > max_len:
+                result[key] = f"<base64 data, {len(item)} chars>"
+            else:
+                result[key] = _shorten_base64_for_log(item, max_len)
+        return result
+    if isinstance(value, list):
+        return [_shorten_base64_for_log(item, max_len) for item in value]
+    if isinstance(value, str) and value.startswith("data:image") and len(value) > max_len:
+        return f"<data image url, {len(value)} chars>"
+    return value
+
+
+def _log_body(label: str, text_or_body: Any) -> None:
+    if isinstance(text_or_body, str):
+        try:
+            text_or_body = json.loads(text_or_body)
+        except Exception:
+            print(f"{label}\n{text_or_body}")
+            return
+    print(
+        f"{label}\n"
+        f"{json.dumps(_shorten_base64_for_log(text_or_body), ensure_ascii=False, indent=2)}"
+    )
+
+
+def build_nano_banana_submit_body(
+    model: str,
+    prompt: str,
+    resolution: str,
+    aspect_ratio: str,
+    images: Optional[List[Image.Image]] = None,
+    enable_grounding: bool = False,
+    thinking_level: Optional[str] = None,
+    request_log_enabled: bool = False,
+    node_label: str = "Nano Banana",
+) -> Dict[str, Any]:
+    def _make_body(image_urls: List[str]) -> Dict[str, Any]:
+        body: Dict[str, Any] = {
+            "model": model,
+            "prompt": prompt,
+            "size": resolution,
+        }
+        if aspect_ratio and aspect_ratio != "智能":
+            body["aspect_ratio"] = aspect_ratio
+        if image_urls:
+            body["images"] = image_urls
+        if enable_grounding:
+            body["google_search"] = True
+        if thinking_level:
+            body["thinking_level"] = thinking_level
+        return body
+
+    working_images = list(images or [])
+    image_urls: List[str] = []
+
+    if working_images:
+        image_urls, working_images, _, _ = _fit_image_data_urls_to_body_limit(
+            working_images,
+            _make_body,
+        )
+
+    body = _make_body(image_urls)
+    body_size = _json_size(body)
+
+    if working_images and body_size > _MAX_BODY_BYTES:
+        raise ValueError(
+            f"Request body exceeds the 20MB limit after compression "
+            f"({body_size / 1_000_000:.2f}MB). Reduce reference image count, "
+            "image complexity, or prompt length."
+        )
+    if not working_images and body_size > _MAX_BODY_BYTES:
+        raise ValueError(
+            f"Request body exceeds the 20MB limit ({body_size / 1_000_000:.2f}MB). "
+            "Shorten the prompt or system instructions."
+        )
+
+    if request_log_enabled:
+        print(
+            f"[{node_label} 异步请求体] {body_size / 1024:.1f}KB\n"
+            f"{json.dumps(_shorten_base64_for_log(body), ensure_ascii=False, indent=2)}"
+        )
+
+    return body
+
+
+async def _interruptible_sleep(
+    seconds: float,
+    check_interrupt: Optional[Callable[[], None]] = None,
+) -> None:
+    elapsed = 0.0
+    while elapsed < seconds:
+        if check_interrupt:
+            check_interrupt()
+        delay = min(_INTERRUPT_STEP, seconds - elapsed)
+        await asyncio.sleep(delay)
+        elapsed += delay
+    if check_interrupt:
+        check_interrupt()
+
+
+def _payload_sources(payload: Dict[str, Any]) -> Iterable[Dict[str, Any]]:
+    queue = [payload]
+    seen = set()
+    while queue:
+        current = queue.pop(0)
+        if not isinstance(current, dict):
+            continue
+        obj_id = id(current)
+        if obj_id in seen:
+            continue
+        seen.add(obj_id)
+        yield current
+        for key in ("data", "result", "response", "output", "task_result", "content"):
+            value = current.get(key)
+            if isinstance(value, dict):
+                queue.append(value)
+
+
+def _extract_task_id(payload: Dict[str, Any]) -> str:
+    for source in _payload_sources(payload):
+        for key in ("task_id", "taskId", "id"):
+            value = source.get(key)
+            if value:
+                return str(value)
+    raise RuntimeError(f"提交响应中未找到 task_id: {payload}")
+
+
+def _extract_status(payload: Dict[str, Any]) -> str:
+    statuses = []
+    for source in _payload_sources(payload):
+        for key in ("status", "task_status", "state", "task_state"):
+            value = source.get(key)
+            if value is not None and str(value).strip():
+                statuses.append(str(value).strip())
+
+    for status in statuses:
+        normalized = status.lower()
+        if normalized in _FAILURE_STATUSES or any(
+            token in normalized for token in ("fail", "error", "reject", "timeout", "cancel")
+        ):
+            return status
+    for status in statuses:
+        if status.lower() in _RUNNING_STATUSES:
+            return status
+    for status in statuses:
+        if status.lower() in _SUCCESS_STATUSES:
+            return status
+    return statuses[0] if statuses else ""
+
+
+def _coerce_progress_fraction(value: Any) -> Optional[float]:
+    if value is None or isinstance(value, bool):
+        return None
+
+    if isinstance(value, (int, float)):
+        progress = float(value)
+    elif isinstance(value, str):
+        text = value.strip()
+        if not text:
+            return None
+        has_percent_suffix = text.endswith("%")
+        if has_percent_suffix:
+            text = text[:-1].strip()
+        try:
+            progress = float(text)
+        except ValueError:
+            return None
+        if has_percent_suffix:
+            progress /= 100.0
+    else:
+        return None
+
+    if progress > 1.0:
+        progress /= 100.0
+    return max(0.0, min(progress, 1.0))
+
+
+def _extract_progress(payload: Dict[str, Any]) -> Optional[float]:
+    for source in _payload_sources(payload):
+        for key in ("progress", "percentage", "percent"):
+            progress = _coerce_progress_fraction(source.get(key))
+            if progress is not None:
+                return progress
+
+        for key in ("progressInfo", "progress_info"):
+            info = source.get(key)
+            if not isinstance(info, dict):
+                continue
+            for field in ("progress", "percentage", "percent"):
+                progress = _coerce_progress_fraction(info.get(field))
+                if progress is not None:
+                    return progress
+
+    return None
+
+
+def _is_failure_status(normalized_status: str) -> bool:
+    return normalized_status in _FAILURE_STATUSES or any(
+        token in normalized_status for token in ("fail", "error", "reject", "timeout", "cancel")
+    )
+
+
+def _extract_error_message(payload: Dict[str, Any]) -> str:
+    for source in _payload_sources(payload):
+        error = source.get("error")
+        if isinstance(error, dict):
+            for key in ("message", "msg", "detail", "reason", "code"):
+                value = error.get(key)
+                if value:
+                    return str(value)
+        elif error:
+            message = extract_structured_error_message(str(error))
+            return message or str(error)
+
+        for key in (
+            "fail_reason",
+            "failure_reason",
+            "task_status_msg",
+            "status_msg",
+            "error_message",
+            "message",
+            "msg",
+            "reason",
+            "detail",
+        ):
+            value = source.get(key)
+            if value:
+                message = extract_structured_error_message(str(value))
+                return message or str(value)
+    return "未知错误"
+
+
+async def _submit_task(
+    session: aiohttp.ClientSession,
+    base_url: str,
+    api_key: str,
+    body: Dict[str, Any],
+    node_label: str,
+    log_body_enabled: bool = False,
+) -> str:
+    url = f"{base_url}{_SUBMIT_ENDPOINT}"
+    timeout = aiohttp.ClientTimeout(total=120, connect=30, sock_read=120)
+    last_status = None
+    last_text = ""
+
+    for attempt in range(DEFAULT_MAX_RETRIES + 1):
+        try:
+            async with session.post(
+                url,
+                headers=_headers(api_key),
+                data=_json_dumps(body).encode("utf-8"),
+                timeout=timeout,
+            ) as resp:
+                text = await resp.text()
+                if log_body_enabled:
+                    _log_body(f"[{node_label} 异步提交响应] HTTP {resp.status}", text)
+                if resp.status not in (200, 201, 202):
+                    last_status = resp.status
+                    last_text = text
+                    if resp.status in RETRYABLE_STATUS_CODES and attempt < DEFAULT_MAX_RETRIES:
+                        friendly = get_friendly_message(resp.status)
+                        delay = _compute_delay(
+                            attempt,
+                            DEFAULT_BASE_DELAY,
+                            DEFAULT_MAX_DELAY,
+                            DEFAULT_BACKOFF_FACTOR,
+                        )
+                        print(f"{node_label}: {friendly} {delay:.1f}s 后重试提交 ({attempt + 1}/{DEFAULT_MAX_RETRIES})...")
+                        await asyncio.sleep(delay)
+                        continue
+                    raise RuntimeError(get_friendly_message(resp.status, text))
+
+                try:
+                    data = json.loads(text)
+                except Exception:
+                    raise RuntimeError(f"提交响应 JSON 解析失败: {text[:500]}") from None
+                task_id = _extract_task_id(data)
+                status = _extract_status(data) or "SUBMITTED"
+                print(f"{node_label}: 异步任务已提交 | task_id={task_id} | status={status}")
+                return task_id
+        except (
+            aiohttp.ClientConnectorError,
+            aiohttp.ClientOSError,
+            aiohttp.ServerDisconnectedError,
+            asyncio.TimeoutError,
+        ) as e:
+            last_text = str(e)
+            if attempt < DEFAULT_MAX_RETRIES:
+                delay = _compute_delay(
+                    attempt,
+                    DEFAULT_BASE_DELAY,
+                    DEFAULT_MAX_DELAY,
+                    DEFAULT_BACKOFF_FACTOR,
+                )
+                print(f"{node_label}: 网络连接失败，{delay:.1f}s 后重试提交 ({attempt + 1}/{DEFAULT_MAX_RETRIES})...")
+                await asyncio.sleep(delay)
+                continue
+            raise RuntimeError(
+                f"网络连接失败，无法连接 {url}: {str(e)}。请切换节点里的网络线路，或检查 VPN/代理/防火墙。"
+            ) from None
+
+    raise RuntimeError(get_friendly_message(last_status or 0, last_text))
+
+
+async def _poll_task(
+    session: aiohttp.ClientSession,
+    base_url: str,
+    api_key: str,
+    task_id: str,
+    node_label: str,
+    check_interrupt: Optional[Callable[[], None]] = None,
+    log_body_enabled: bool = False,
+    progress_callback: Optional[Callable[[float], None]] = None,
+) -> Dict[str, Any]:
+    url = f"{base_url}{_TASK_ENDPOINT.format(task_id=task_id)}"
+    start_time = time.time()
+    last_poll_at = start_time
+    poll_count = 0
+
+    while True:
+        if check_interrupt:
+            check_interrupt()
+
+        if poll_count < len(_POLL_SCHEDULE):
+            next_poll_at = start_time + _POLL_SCHEDULE[poll_count]
+        else:
+            next_poll_at = last_poll_at + _POLL_INTERVAL
+
+        sleep_time = next_poll_at - time.time()
+        if sleep_time > 0:
+            await _interruptible_sleep(sleep_time, check_interrupt=check_interrupt)
+
+        last_poll_at = time.time()
+        elapsed = last_poll_at - start_time
+        if elapsed > _MAX_WAIT_SECONDS:
+            raise RuntimeError(f"任务 {task_id} 超时（>{int(_MAX_WAIT_SECONDS)}秒），请稍后用 task_id 查询结果")
+
+        poll_count += 1
+        async with session.get(url, headers=_headers(api_key)) as resp:
+            text = await resp.text()
+            if log_body_enabled:
+                _log_body(f"[{node_label} 任务查询响应 #{poll_count}] HTTP {resp.status}", text)
+            if resp.status != 200:
+                raise RuntimeError(get_friendly_message(resp.status, text))
+            try:
+                payload = json.loads(text)
+            except Exception:
+                raise RuntimeError(f"任务查询响应 JSON 解析失败: {text[:500]}") from None
+
+        status = _extract_status(payload) or "UNKNOWN"
+        normalized = status.lower()
+        progress = _extract_progress(payload)
+        is_failure = _is_failure_status(normalized)
+        if _POLL_LOG_ENABLED:
+            progress_text = ""
+            if progress is not None and not is_failure:
+                displayed_progress = 1.0 if normalized in _SUCCESS_STATUSES else min(progress, _RUNNING_PROGRESS_MAX)
+                progress_text = f" | progress={displayed_progress * 100:.0f}%"
+            print(f"{node_label}: 查询任务 #{poll_count} | task_id={task_id} | status={status}{progress_text}")
+
+        if normalized in _SUCCESS_STATUSES:
+            if progress_callback:
+                progress_callback(1.0)
+            return payload
+        if is_failure:
+            raise RuntimeError(f"任务失败: {_extract_error_message(payload)}")
+        if normalized not in _RUNNING_STATUSES:
+            raise RuntimeError(f"未知任务状态 {status}: {payload}")
+        if progress_callback and progress is not None:
+            progress_callback(min(progress, _RUNNING_PROGRESS_MAX))
+
+
+async def _image_from_url_or_data(
+    value: str,
+    session: aiohttp.ClientSession,
+) -> Optional[Image.Image]:
+    if not value:
+        return None
+    if value.startswith("data:image"):
+        try:
+            _, b64_data = value.split(",", 1)
+            return Image.open(BytesIO(base64.b64decode(b64_data))).convert("RGB")
+        except Exception as e:
+            raise RuntimeError(f"data URL 图片解码失败: {e}") from None
+    if value.startswith("http"):
+        async with session.get(value, allow_redirects=True) as resp:
+            if resp.status != 200:
+                raise RuntimeError(f"图片下载失败 ({resp.status}): {value}")
+            img_bytes = await resp.read()
+        return Image.open(BytesIO(img_bytes)).convert("RGB")
+    return None
+
+
+async def _parse_direct_images(
+    payload: Dict[str, Any],
+    session: aiohttp.ClientSession,
+) -> List[Image.Image]:
+    images: List[Image.Image] = []
+
+    async def _try_item(item: Any) -> None:
+        if isinstance(item, str):
+            img = await _image_from_url_or_data(item, session)
+            if img:
+                images.append(img)
+            return
+        if not isinstance(item, dict):
+            return
+
+        for key in ("url", "image_url", "result_url", "download_url"):
+            img = await _image_from_url_or_data(str(item.get(key) or ""), session)
+            if img:
+                images.append(img)
+                return
+
+        b64_data = item.get("b64_json") or item.get("base64") or item.get("image_base64")
+        if b64_data:
+            images.append(Image.open(BytesIO(base64.b64decode(str(b64_data)))).convert("RGB"))
+            return
+
+        for inline_key in ("inline_data", "inlineData"):
+            inline = item.get(inline_key)
+            if isinstance(inline, dict) and inline.get("data"):
+                images.append(Image.open(BytesIO(base64.b64decode(str(inline["data"])))).convert("RGB"))
+                return
+
+    for source in _payload_sources(payload):
+        for key in ("image_url", "result_url", "url", "download_url"):
+            img = await _image_from_url_or_data(str(source.get(key) or ""), session)
+            if img:
+                images.append(img)
+
+        for key in ("images", "output_images", "outputs"):
+            value = source.get(key)
+            if isinstance(value, list):
+                for item in value:
+                    await _try_item(item)
+            elif value:
+                await _try_item(value)
+
+    return images
+
+
+async def _parse_task_images(
+    task_payload: Dict[str, Any],
+    session: aiohttp.ClientSession,
+    api_key: str,
+) -> List[Image.Image]:
+    direct_images = await _parse_direct_images(task_payload, session)
+    if direct_images:
+        return direct_images
+
+    client = GeminiAPIClient(api_key=api_key)
+    last_error = None
+    for source in _payload_sources(task_payload):
+        if "candidates" not in source:
+            continue
+        try:
+            images, _ = await client.parse_response_async(source, session=session)
+            if images:
+                return [img.convert("RGB") for img in images]
+        except Exception as e:
+            last_error = e
+
+    if last_error is not None:
+        raise RuntimeError(str(last_error)) from None
+    raise RuntimeError(f"任务成功但未找到图片结果: {task_payload}")
+
+
+async def generate_nano_banana_async(
+    session: aiohttp.ClientSession,
+    base_url: str,
+    api_key: str,
+    prompt: str,
+    model: str,
+    resolution: str,
+    aspect_ratio: str,
+    images: Optional[List[Image.Image]] = None,
+    enable_grounding: bool = False,
+    thinking_level: Optional[str] = None,
+    node_label: str = "Nano Banana",
+    request_log_enabled: bool = False,
+    check_interrupt: Optional[Callable[[], None]] = None,
+    progress_callback: Optional[Callable[[float], None]] = None,
+) -> Tuple[List[Image.Image], Dict[str, Any]]:
+    if check_interrupt:
+        check_interrupt()
+
+    body = build_nano_banana_submit_body(
+        model=model,
+        prompt=prompt,
+        resolution=resolution,
+        aspect_ratio=aspect_ratio,
+        images=images,
+        enable_grounding=enable_grounding,
+        thinking_level=thinking_level,
+        request_log_enabled=request_log_enabled,
+        node_label=node_label,
+    )
+
+    task_start = time.time()
+    task_id = await _submit_task(
+        session,
+        base_url,
+        api_key,
+        body,
+        node_label,
+        log_body_enabled=request_log_enabled,
+    )
+    task_payload = await _poll_task(
+        session,
+        base_url,
+        api_key,
+        task_id,
+        node_label,
+        check_interrupt=check_interrupt,
+        log_body_enabled=request_log_enabled,
+        progress_callback=progress_callback,
+    )
+    task_done = time.time()
+
+    parse_start = time.time()
+    images_list = await _parse_task_images(task_payload, session, api_key)
+    parse_done = time.time()
+
+    return images_list, {
+        "task_id": task_id,
+        "task_ms": (task_done - task_start) * 1000,
+        "parse_ms": (parse_done - parse_start) * 1000,
+        "request_bytes": _json_size(body),
+    }
@@ -0,0 +1,145 @@
+"""
+R2 文件上传工具（通过 o1key 后端预签名接口）
+- 插件内零 R2 凭证，仅使用用户的 O1KEY_API_KEY
+- 流程：请求预签名 URL → PUT 直传 R2 → 返回公网 URL
+"""
+
+import io
+import os
+import uuid
+
+import aiohttp
+
+from .config import get_api_key_or_raise, get_api_base_url
+
+
+async def _presign(filename: str, content_type: str) -> tuple:
+    """向 o1key 后端请求预签名 URL，返回 (upload_url, public_url)"""
+    api_key = get_api_key_or_raise()
+    base_url = get_api_base_url()
+
+    connector = aiohttp.TCPConnector(ssl=False)
+    async with aiohttp.ClientSession(connector=connector) as session:
+        async with session.post(
+            f"{base_url}/v1/storage/presign",
+            headers={"Authorization": f"Bearer {api_key}"},
+            json={"filename": filename, "content_type": content_type},
+            timeout=aiohttp.ClientTimeout(total=10),
+        ) as resp:
+            if resp.status != 200:
+                text = await resp.text()
+                raise RuntimeError(f"预签名请求失败 ({resp.status}): {text}")
+            data = await resp.json()
+
+    return data["upload_url"], data["public_url"]
+
+
+async def _put_upload(upload_url: str, data: bytes, content_type: str):
+    """用预签名 URL 直传文件到 R2（不带 Authorization）"""
+    connector = aiohttp.TCPConnector(ssl=False)
+    async with aiohttp.ClientSession(connector=connector) as session:
+        async with session.put(
+            upload_url,
+            data=data,
+            headers={"Content-Type": content_type},
+            timeout=aiohttp.ClientTimeout(total=120),
+        ) as resp:
+            if resp.status not in (200, 204):
+                text = await resp.text()
+                raise RuntimeError(f"文件上传失败 ({resp.status}): {text}")
+
+
+async def upload_image(pil_image) -> str:
+    """
+    接受 PIL Image 对象，编码为 PNG 上传到 R2，返回公网 URL。
+    """
+    import io as _io
+    buf = _io.BytesIO()
+    pil_image.save(buf, format="PNG")
+    data = buf.getvalue()
+    filename = f"{uuid.uuid4()}.png"
+
+    upload_url, public_url = await _presign(filename, "image/png")
+    await _put_upload(upload_url, data, "image/png")
+
+    print(f"[R2] 图片已上传: {public_url}")
+    return public_url
+
+
+async def upload_video(video) -> str:
+    """
+    接受 ComfyUI VIDEO 对象，上传到 R2，返回公网 URL。
+    支持 mp4 / mov 格式。
+    """
+    source = video.get_stream_source()
+
+    if isinstance(source, io.BytesIO):
+        source.seek(0)
+        data = source.read()
+        ext = "mp4"
+    else:
+        video_path = source
+        if not video_path or not os.path.isfile(video_path):
+            raise ValueError(f"无法获取参考视频文件路径（当前路径：{video_path}）")
+        ext = os.path.splitext(video_path)[1].lower().lstrip(".")
+        if ext not in ("mp4", "mov"):
+            raise ValueError(f"参考视频格式须为 mp4 或 mov，当前为 .{ext}")
+        with open(video_path, "rb") as f:
+            data = f.read()
+
+    content_type = "video/mp4" if ext == "mp4" else "video/quicktime"
+    filename = f"{uuid.uuid4()}.{ext}"
+
+    upload_url, public_url = await _presign(filename, content_type)
+    await _put_upload(upload_url, data, content_type)
+
+    print(f"[R2] 视频已上传: {public_url}")
+    return public_url
+
+
+async def upload_audio(audio) -> str:
+    """
+    接受 ComfyUI AUDIO dict（waveform tensor + sample_rate），
+    编码为 WAV 后上传到 R2，返回公网 URL。
+    """
+    import struct
+    import numpy as np
+
+    waveform = audio["waveform"]      # shape: [B, C, N] or [C, N]
+    sample_rate = int(audio["sample_rate"])
+
+    if waveform.dim() == 3:
+        waveform = waveform[0]
+
+    wav_np = waveform.cpu().numpy()
+    if wav_np.ndim == 2:
+        wav_np = wav_np.mean(axis=0)
+    wav_np = np.clip(wav_np, -1.0, 1.0)
+    pcm = (wav_np * 32767).astype(np.int16)
+
+    num_samples = len(pcm)
+    num_channels = 1
+    bits_per_sample = 16
+    byte_rate = sample_rate * num_channels * bits_per_sample // 8
+    block_align = num_channels * bits_per_sample // 8
+    data_size = num_samples * block_align
+
+    buf = io.BytesIO()
+    buf.write(b"RIFF")
+    buf.write(struct.pack("<I", 36 + data_size))
+    buf.write(b"WAVE")
+    buf.write(b"fmt ")
+    buf.write(struct.pack("<IHHIIHH", 16, 1, num_channels, sample_rate,
+                          byte_rate, block_align, bits_per_sample))
+    buf.write(b"data")
+    buf.write(struct.pack("<I", data_size))
+    buf.write(pcm.tobytes())
+
+    data = buf.getvalue()
+    filename = f"{uuid.uuid4()}.wav"
+
+    upload_url, public_url = await _presign(filename, "audio/wav")
+    await _put_upload(upload_url, data, "audio/wav")
+
+    print(f"[R2] 音频已上传: {public_url}")
+    return public_url
@@ -0,0 +1,73 @@
+"""
+背景移除工具模块
+基于 rembg 库实现，支持 CPU 推理
+"""
+
+import os
+import numpy as np
+import torch
+from PIL import Image
+
+_session = None
+
+
+def _get_session():
+    """懒加载 rembg session，避免启动时加载模型"""
+    global _session
+    if _session is None:
+        try:
+            import folder_paths
+            models_dir = os.path.join(folder_paths.models_dir, "rembg")
+            os.makedirs(models_dir, exist_ok=True)
+            os.environ["U2NET_HOME"] = models_dir
+        except Exception:
+            pass
+        try:
+            from rembg import new_session
+            _session = new_session("isnet-general-use")
+            print("[o1key] rembg 模型加载完成 (isnet-general-use)")
+        except ImportError:
+            raise RuntimeError(
+                "未安装 rembg，请执行: pip install rembg[cpu]>=2.0.50"
+            )
+    return _session
+
+
+def remove_background_pil(image: Image.Image) -> Image.Image:
+    """
+    移除 PIL Image 背景，返回 RGBA 图像（背景透明）
+    """
+    from rembg import remove
+    session = _get_session()
+    result = remove(image, session=session)
+    return result.convert("RGBA")
+
+
+def remove_background_tensor(tensor: torch.Tensor) -> torch.Tensor:
+    """
+    移除 ComfyUI IMAGE tensor 的背景
+    输入: [B, H, W, C] (3或4通道)
+    输出: [B, H, W, 4] RGBA tensor
+    """
+    from rembg import remove
+    session = _get_session()
+
+    results = []
+    batch_size = tensor.shape[0]
+
+    for i in range(batch_size):
+        frame = tensor[i]  # [H, W, C]
+        arr = (frame.cpu().numpy() * 255).clip(0, 255).astype(np.uint8)
+
+        if arr.shape[2] == 4:
+            pil_img = Image.fromarray(arr, mode="RGBA")
+        else:
+            pil_img = Image.fromarray(arr, mode="RGB")
+
+        result = remove(pil_img, session=session)
+        result_rgba = result.convert("RGBA")
+
+        result_arr = np.array(result_rgba).astype(np.float32) / 255.0
+        results.append(torch.from_numpy(result_arr))
+
+    return torch.stack(results, dim=0)
@@ -39,12 +39,15 @@ def check_for_updates() -> bool:
        if not os.path.exists(git_dir):
            return False
        
-        # 执行 git fetch
+        # 执行 git fetch（禁止弹出认证弹框，失败时静默处理）
+        env = os.environ.copy()
+        env['GIT_TERMINAL_PROMPT'] = '0'
        subprocess.run(
            ['git', 'fetch', 'origin'],
            cwd=plugin_dir,
            capture_output=True,
-            timeout=10
+            timeout=10,
+            env=env
        )
        
        # 检查本地和远程版本
@@ -68,16 +71,78 @@ def check_for_updates() -> bool:
        return False


+def get_update_changelog() -> list:
+    """从远程 CHANGELOG.md 最新版本块中提取更新内容（最多5条）"""
+    try:
+        plugin_dir = os.path.dirname(os.path.dirname(__file__))
+        result = subprocess.run(
+            ['git', 'show', 'origin/main:CHANGELOG.md'],
+            cwd=plugin_dir,
+            capture_output=True,
+            text=True,
+            encoding='utf-8'
+        )
+        lines = result.stdout.splitlines()
+
+        in_block = False
+        items = []
+        for line in lines:
+            if line.startswith('## [') and not line.startswith('## [Unreleased]'):
+                if in_block:
+                    break
+                in_block = True
+                continue
+            if in_block:
+                stripped = line.strip()
+                if stripped and not stripped.startswith('#') and not stripped.startswith('---'):
+                    text = stripped.lstrip('- ').replace('**', '').strip()
+                    if text and len(text) > 3:
+                        items.append(text)
+                if len(items) >= 5:
+                    break
+
+        return items
+    except Exception:
+        return []
+
+
+def notify_new_version():
+    """检测到新版本时，推送蓝色更新通知弹框"""
+    changelog = get_update_changelog()
+
+    try:
+        import threading
+        from server import PromptServer
+
+        def _send():
+            try:
+                PromptServer.instance.send_sync(
+                    "o1key.new_version",
+                    {"changelog": changelog}
+                )
+            except Exception:
+                pass
+
+        threading.Timer(3.0, _send).start()
+    except Exception:
+        pass
+
+
 def notify_update_available():
-    """通知用户有更新可用"""
-    current_version = get_current_version()
-    version_str = f" (当前版本: {current_version})" if current_version else ""
-    
-    print("\n" + "="*60)
-    print(f"🎉 Comfyui_o1key 有新版本可用{version_str}")
-    print("="*60)
-    print("更新方法：")
-    print("  Windows: 双击运行 update.bat")
-    print("  Linux/Mac: 运行 ./update.sh")
-    print("或手动执行: git pull origin main")
-    print("="*60 + "\n")
+    """通知用户有更新可用（前端弹窗）"""
+    try:
+        import threading
+        from server import PromptServer
+
+        def _send():
+            try:
+                PromptServer.instance.send_sync(
+                    "o1key.update_available",
+                    {"message": "欢迎使用o1key工作流，祝您马年，马上有福，马上有钱，马到成功！！！"}
+                )
+            except Exception:
+                pass
+
+        threading.Timer(3.0, _send).start()
+    except Exception:
+        pass
@@ -0,0 +1,68 @@
+"""Safely fast-forward a Git installation of this node package."""
+
+import os
+import subprocess
+from pathlib import Path
+
+
+PLUGIN_DIR = Path(__file__).resolve().parent.parent
+
+
+class UpdateError(Exception):
+    pass
+
+
+def _git(*args, timeout=60, check=True):
+    env = os.environ.copy()
+    env["GIT_TERMINAL_PROMPT"] = "0"
+    env["GCM_INTERACTIVE"] = "Never"
+    try:
+        result = subprocess.run(
+            ["git", *args],
+            cwd=PLUGIN_DIR,
+            env=env,
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            timeout=timeout,
+        )
+    except FileNotFoundError as exc:
+        raise UpdateError("未找到 Git，请先安装 Git。") from exc
+    except subprocess.TimeoutExpired as exc:
+        raise UpdateError("Git 操作超时，请检查网络后重试。") from exc
+    if check and result.returncode:
+        detail = (result.stderr or result.stdout).strip().splitlines()
+        raise UpdateError(detail[-1] if detail else "Git 操作失败。")
+    return result
+
+
+def update_package():
+    """Update origin/main without discarding local changes or switching branches."""
+    if not (PLUGIN_DIR / ".git").exists():
+        raise UpdateError("当前节点包不是 Git 安装。请通过 Git 安装后再使用界面更新。")
+
+    branch = _git("symbolic-ref", "--quiet", "--short", "HEAD", check=False)
+    if branch.returncode or branch.stdout.strip() != "main":
+        raise UpdateError("当前不在 main 分支，请手动检查分支后更新。")
+
+    if _git("status", "--porcelain", "--untracked-files=no").stdout.strip():
+        raise UpdateError("节点包有本地修改，请先保存或处理修改后再更新。")
+
+    old_commit = _git("rev-parse", "HEAD").stdout.strip()
+    old_requirements = _git("show", "HEAD:requirements.txt", check=False).stdout
+    _git("fetch", "origin", "main")
+    new_commit = _git("rev-parse", "FETCH_HEAD").stdout.strip()
+    if old_commit == new_commit:
+        return {"updated": False, "version": old_commit[:7], "requirements_changed": False}
+
+    if _git("merge-base", "--is-ancestor", "HEAD", "FETCH_HEAD", check=False).returncode:
+        raise UpdateError("本地与 origin/main 已分叉，无法安全快进。请手动处理。")
+
+    _git("merge", "--ff-only", "FETCH_HEAD")
+    requirements_changed = old_requirements != (PLUGIN_DIR / "requirements.txt").read_text(encoding="utf-8")
+    return {
+        "updated": True,
+        "version": new_commit[:7],
+        "requirements_changed": requirements_changed,
+    }
@@ -0,0 +1,181 @@
+import asyncio
+from typing import Any, Dict
+
+
+try:
+    from comfy.model_management import processing_interrupted, InterruptProcessingException
+    INTERRUPT_AVAILABLE = True
+except Exception:
+    INTERRUPT_AVAILABLE = False
+    processing_interrupted = lambda: False
+
+    class InterruptProcessingException(Exception):
+        pass
+
+
+SUCCESS_STATUSES = {"succeed", "succeeded", "success", "completed", "done", "finished"}
+FAILURE_STATUSES = {
+    "fail",
+    "failed",
+    "failure",
+    "error",
+    "expired",
+    "timeout",
+    "timed_out",
+    "cancel",
+    "canceled",
+    "cancelled",
+    "rejected",
+}
+
+
+def check_interrupt() -> None:
+    if INTERRUPT_AVAILABLE and processing_interrupted():
+        raise InterruptProcessingException()
+
+
+async def interruptible_sleep(seconds: float, step: float = 0.2) -> None:
+    elapsed = 0.0
+    while elapsed < seconds:
+        check_interrupt()
+        delay = min(step, seconds - elapsed)
+        await asyncio.sleep(delay)
+        elapsed += delay
+    check_interrupt()
+
+
+async def run_with_interrupt(coro, step: float = 0.2):
+    task = asyncio.ensure_future(coro)
+    try:
+        while not task.done():
+            check_interrupt()
+            await asyncio.wait({task}, timeout=step)
+        check_interrupt()
+        return await task
+    except InterruptProcessingException:
+        task.cancel()
+        try:
+            await task
+        except BaseException:
+            pass
+        raise
+
+
+def _as_dict(value: Any) -> Dict[str, Any]:
+    return value if isinstance(value, dict) else {}
+
+
+def _nested_payloads(payload: Dict[str, Any]):
+    root = _as_dict(payload)
+    data = _as_dict(root.get("data"))
+    inner = _as_dict(data.get("data"))
+    return root, data, inner
+
+
+def extract_status(payload: Dict[str, Any]) -> str:
+    root, data, inner = _nested_payloads(payload)
+    keys = ("status", "task_status", "state", "task_state")
+    statuses = []
+    for source in (data, inner, root):
+        for key in keys:
+            value = source.get(key)
+            if value is not None and str(value).strip():
+                statuses.append(str(value).strip().lower())
+    for status in statuses:
+        if status in FAILURE_STATUSES or any(
+            token in status for token in ("fail", "error", "reject", "timeout", "cancel")
+        ):
+            return status
+    for status in statuses:
+        if status in SUCCESS_STATUSES:
+            return status
+    return statuses[0] if statuses else ""
+
+
+def extract_progress(payload: Dict[str, Any]) -> int:
+    root, data, inner = _nested_payloads(payload)
+    for source in (data, inner, root):
+        value = source.get("progress")
+        if value is None:
+            continue
+        try:
+            return max(0, min(100, int(float(str(value).strip().rstrip("%")))))
+        except (TypeError, ValueError):
+            return 0
+    return 0
+
+
+def extract_error_message(payload: Dict[str, Any], default: str = "未知错误") -> str:
+    root, data, inner = _nested_payloads(payload)
+    keys = (
+        "fail_reason",
+        "failure_reason",
+        "task_status_msg",
+        "status_msg",
+        "error_message",
+        "message",
+        "msg",
+        "reason",
+        "detail",
+        "details",
+    )
+    for source in (data, inner, root):
+        error = source.get("error")
+        if isinstance(error, dict):
+            for key in ("message", "msg", "detail", "reason", "code"):
+                value = error.get(key)
+                if value:
+                    return str(value)
+        elif error:
+            return str(error)
+
+        for key in keys:
+            value = source.get(key)
+            if value:
+                return str(value)
+    return default
+
+
+def extract_video_url(payload: Dict[str, Any]) -> str | None:
+    root, data, inner = _nested_payloads(payload)
+    for source in (data, inner, root):
+        for key in ("video_url", "result_url", "url", "download_url"):
+            value = source.get(key)
+            if value:
+                return str(value)
+
+        result = _as_dict(source.get("result"))
+        for key in ("video_url", "result_url", "url", "download_url"):
+            value = result.get(key)
+            if value:
+                return str(value)
+
+        content = _as_dict(source.get("content"))
+        value = content.get("video_url") or content.get("url")
+        if value:
+            return str(value)
+
+        task_result = _as_dict(source.get("task_result"))
+        videos = task_result.get("videos")
+        if isinstance(videos, list) and videos:
+            first = _as_dict(videos[0])
+            value = first.get("url") or first.get("video_url")
+            if value:
+                return str(value)
+    return None
+
+
+def is_success_status(status: str) -> bool:
+    return status in SUCCESS_STATUSES
+
+
+def is_failure_status(status: str, payload: Dict[str, Any] | None = None) -> bool:
+    if status in FAILURE_STATUSES:
+        return True
+    if any(token in status for token in ("fail", "error", "reject", "timeout", "cancel")):
+        return True
+    if payload is None:
+        return False
+    root, data, inner = _nested_payloads(payload)
+    failure_keys = ("error", "fail_reason", "failure_reason", "task_status_msg", "error_message")
+    return any(any(source.get(key) for key in failure_keys) for source in (data, inner, root))
@@ -1 +1 @@
-v1.10.0
+v1.10.3
@@ -0,0 +1,108 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+app.registerExtension({
+    name: "o1key.assetToggle",
+    settings: [
+        {
+            id: "o1key.AssetSave",
+            name: "资产保存",
+            tooltip: "持久性保存生图记录",
+            type: "boolean",
+            defaultValue: true,
+        },
+    ],
+    async init() {
+        const enabled = () => {
+            try {
+                return app.ui.settings.getSettingValue("o1key.AssetSave", true);
+            } catch {
+                return true;
+            }
+        };
+
+        // --- Patch getHistory: 合并真实 jobs 与持久化历史 ---
+        const _origGetHistory = api.getHistory.bind(api);
+        api.getHistory = async function (maxItems = 200, opts = {}) {
+            const real = await _origGetHistory(maxItems, opts);
+            if (!enabled()) return real;
+            try {
+                const offset = opts?.offset || 0;
+                const resp = await fetch(
+                    `/o1key/output_history?limit=${maxItems}&offset=${offset}`
+                );
+                if (!resp.ok) return real;
+                const data = await resp.json();
+                const persisted = data.jobs || [];
+                if (!persisted.length) return real;
+                if (!real || !Array.isArray(real) || !real.length) {
+                    // 仅持久化数据时也补充 priority
+                    const t = data.pagination?.total || persisted.length;
+                    return persisted.map((j, i) => ({
+                        ...j,
+                        priority: j.priority ?? t - i,
+                    }));
+                }
+                // 合并去重：以 id 为 key，真实 jobs 优先
+                const seen = new Set(real.map((j) => j.id));
+                const merged = [...real];
+                for (const job of persisted) {
+                    if (!seen.has(job.id)) {
+                        merged.push(job);
+                    }
+                }
+                // 按时间倒序
+                merged.sort(
+                    (a, b) => (b.create_time || 0) - (a.create_time || 0)
+                );
+                const result = merged.slice(0, maxItems);
+                // 补充 priority 字段（队列面板依赖此字段排序）
+                const total = data.pagination?.total || result.length;
+                for (let idx = 0; idx < result.length; idx++) {
+                    if (result[idx].priority == null) {
+                        result[idx] = {
+                            ...result[idx],
+                            priority: total - idx,
+                        };
+                    }
+                }
+                return result;
+            } catch (e) {
+                return real;
+            }
+        };
+
+        // --- Patch deleteItem: 同时删除 o1key 持久化记录和文件 ---
+        const _origDeleteItem = api.deleteItem.bind(api);
+        api.deleteItem = async function (type, id) {
+            const result = await _origDeleteItem(type, id);
+            if (type === "history" && enabled()) {
+                try {
+                    await fetch("/o1key/delete_history", {
+                        method: "POST",
+                        headers: { "Content-Type": "application/json" },
+                        body: JSON.stringify({ delete: [id] }),
+                    });
+                } catch (e) {}
+            }
+            return result;
+        };
+
+        // --- Patch getJobDetail: 真实 API 失败时回退到本地路由 ---
+        const _origGetJobDetail = api.getJobDetail.bind(api);
+        api.getJobDetail = async function (jobId) {
+            const real = await _origGetJobDetail(jobId);
+            if (real) return real;
+            if (!enabled()) return undefined;
+            try {
+                const resp = await fetch(
+                    `/o1key/job_detail/${encodeURIComponent(jobId)}`
+                );
+                if (!resp.ok) return undefined;
+                return await resp.json();
+            } catch (e) {
+                return undefined;
+            }
+        };
+    },
+});
@@ -0,0 +1,41 @@
+import { app } from "../../../scripts/app.js";
+
+app.registerExtension({
+    name: "o1key.dotGrid",
+    async setup() {
+        function createBlackTile() {
+            const size = 64;
+            const c = document.createElement("canvas");
+            c.width = size;
+            c.height = size;
+            const ctx = c.getContext("2d");
+            ctx.fillStyle = "#1a1a1a";
+            ctx.fillRect(0, 0, size, size);
+            return c;
+        }
+
+        // Hook immediately so the first draw already uses our tile
+        const orig = LGraphCanvas.prototype.drawBackCanvas;
+        LGraphCanvas.prototype.drawBackCanvas = function () {
+            if (!this._pattern || !this._pattern_img) {
+                const ctx = this.bgcanvas?.getContext("2d");
+                if (ctx) {
+                    const t = createBlackTile();
+                    this._pattern = ctx.createPattern(t, "repeat");
+                    this._pattern_img = t;
+                }
+            }
+            return orig.apply(this, arguments);
+        };
+
+        // Also apply to current canvas instance if already exists
+        const canvas = app.canvas;
+        if (canvas?.bgcanvas) {
+            const bgCtx = canvas.bgcanvas.getContext("2d");
+            const tile = createBlackTile();
+            canvas._pattern = bgCtx.createPattern(tile, "repeat");
+            canvas._pattern_img = tile;
+            canvas.draw(true, true);
+        }
+    },
+});
@@ -0,0 +1,122 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+// 上传单个文件到 ComfyUI input 目录，返回服务端绝对路径
+async function uploadFile(file) {
+    const formData = new FormData();
+    formData.append("image", file, file.name);
+    const resp = await api.fetchApi("/upload/image", { method: "POST", body: formData });
+    if (!resp.ok) throw new Error(`上传失败: ${file.name}`);
+    const data = await resp.json();
+    const inputDir = await getInputDir();
+    // 拼成绝对路径（Windows 用反斜杠也可以，用正斜杠 Python 也认）
+    return inputDir ? inputDir.replace(/\\/g, "/") + "/" + data.name : data.name;
+}
+
+// 获取 ComfyUI input 目录绝对路径（缓存）
+let _inputDir = null;
+async function getInputDir() {
+    if (_inputDir !== null) return _inputDir;
+    try {
+        const resp = await api.fetchApi("/o1key/input_dir");
+        if (resp.ok) _inputDir = (await resp.json()).path;
+        else _inputDir = "";
+    } catch { _inputDir = ""; }
+    return _inputDir;
+}
+
+// 创建一个"选择文件"按钮，点击后弹出文件选择框
+// onPaths(paths: string[]) 回调拿到上传后的路径列表
+function makeUploadButton(label, accept, multiple, onPaths) {
+    const btn = document.createElement("button");
+    btn.textContent = label;
+    btn.style.cssText =
+        "width:100%;padding:4px 8px;cursor:pointer;margin-top:2px;" +
+        "background:#3a5a3a;color:#ddd;border:1px solid #666;" +
+        "border-radius:4px;font-size:12px;";
+
+    const fileInput = document.createElement("input");
+    fileInput.type = "file";
+    fileInput.multiple = multiple;
+    fileInput.accept = accept;
+    fileInput.style.display = "none";
+    document.body.appendChild(fileInput);
+
+    btn.addEventListener("click", () => fileInput.click());
+
+    fileInput.addEventListener("change", async () => {
+        const files = Array.from(fileInput.files);
+        if (!files.length) return;
+        btn.textContent = "⏳ 上传中...";
+        btn.disabled = true;
+        try {
+            const paths = [];
+            for (const f of files) paths.push(await uploadFile(f));
+            onPaths(paths);
+            btn.textContent = `✅ 已上传 ${files.length} 个`;
+            setTimeout(() => { btn.textContent = label; }, 2000);
+        } catch (e) {
+            console.error("[o1key fileUpload]", e);
+            btn.textContent = "❌ 上传失败";
+            setTimeout(() => { btn.textContent = label; }, 2000);
+        } finally {
+            btn.disabled = false;
+            fileInput.value = "";
+        }
+    });
+
+    return btn;
+}
+
+const ACCEPT = ".pdf,.txt,.md,.csv,.json,.py,.js,.ts,.html,.xml,.docx,.xlsx,.pptx,.zip,.wav,.mp3,.png,.jpg,.jpeg,.webp";
+
+app.registerExtension({
+    name: "o1key.fileUpload",
+
+    async beforeRegisterNodeDef(nodeType, nodeData) {
+        if (nodeData.name !== "LoadFile") return;
+
+        const origCreated = nodeType.prototype.onNodeCreated;
+        nodeType.prototype.onNodeCreated = function () {
+            origCreated?.call(this);
+
+            const singleWidget = this.widgets?.find(w => w.name === "单文件路径");
+            const folderWidget = this.widgets?.find(w => w.name === "文件夹路径");
+
+            // "单文件路径"下方加按钮（支持多选，追加路径）
+            if (singleWidget) {
+                const btn = makeUploadButton("📂 选择文件（可多选）", ACCEPT, true, (paths) => {
+                    const existing = singleWidget.value?.trim();
+                    singleWidget.value = existing
+                        ? existing + ", " + paths.join(", ")
+                        : paths.join(", ");
+                    singleWidget.callback?.(singleWidget.value);
+                    app.graph.setDirtyCanvas(true);
+                });
+                this.addDOMWidget("upload_single_btn", "btn", btn, {
+                    getValue() { return null; },
+                    setValue() {},
+                });
+            }
+
+            // 清空按钮：同时清空单文件路径和文件夹路径
+            if (singleWidget || folderWidget) {
+                const clearBtn = document.createElement("button");
+                clearBtn.textContent = "🗑 清空文件路径";
+                clearBtn.style.cssText =
+                    "width:100%;padding:4px 8px;cursor:pointer;margin-top:2px;" +
+                    "background:#5a3a3a;color:#ddd;border:1px solid #666;" +
+                    "border-radius:4px;font-size:12px;";
+                clearBtn.addEventListener("click", () => {
+                    if (singleWidget) { singleWidget.value = ""; singleWidget.callback?.(""); }
+                    if (folderWidget) { folderWidget.value = ""; folderWidget.callback?.(""); }
+                    app.graph.setDirtyCanvas(true);
+                });
+                this.addDOMWidget("clear_paths_btn", "btn", clearBtn, {
+                    getValue() { return null; },
+                    setValue() {},
+                });
+            }
+        };
+    },
+});
@@ -0,0 +1,221 @@
+import { app } from "../../../scripts/app.js";
+
+app.registerExtension({
+    name: "o1key.hideSidebarItems",
+    async setup() {
+        const hide = () => {
+            // 隐藏侧边栏的"说明"、"应用"、"模型"、"节点"、"模板"按钮
+            const hiddenLabels = ["说明", "帮助", "help", "应用", "apps", "模型", "models", "节点", "nodes", "模板", "templates", "template"];
+            document.querySelectorAll(".side-bar-button, [class*='sidebar'] button, .p-togglebutton").forEach(btn => {
+                const label = [
+                    btn.getAttribute("aria-label"),
+                    btn.getAttribute("title"),
+                    btn.getAttribute("data-title"),
+                    btn.getAttribute("data-label"),
+                    btn.textContent,
+                ].filter(Boolean).join(" ").toLowerCase();
+                if (hiddenLabels.some(k => label.includes(k))) {
+                    btn.style.display = "none";
+                }
+            });
+            // 隐藏左上角下拉菜单中的"帮助"项
+            document.querySelectorAll(".p-menuitem, .p-menu-item, [class*='menu'] li, [class*='Menu'] li").forEach(item => {
+                const text = item.textContent || "";
+                if (text.trim() === "帮助" || text.trim() === "Help") {
+                    item.style.display = "none";
+                }
+            });
+            // 隐藏登录/注册弹框（Google/Github 登录对话框）
+            document.querySelectorAll("[class*='dialog'], [class*='Dialog'], [class*='modal'], [class*='Modal']").forEach(dialog => {
+                const text = dialog.textContent || "";
+                if ((text.includes("Google") || text.includes("Github")) &&
+                    (text.includes("登录") || text.includes("注册"))) {
+                    dialog.style.display = "none";
+                    const mask = dialog.previousElementSibling;
+                    if (mask && mask.className && mask.className.includes("mask")) {
+                        mask.style.display = "none";
+                    }
+                }
+            });
+            // 在右侧内容区隐藏"登录/注册"按钮并注入 API Key
+            injectApiKeyPanel();
+        };
+
+        async function injectApiKeyPanel() {
+            // 找到右侧内容区中包含"我的用户设置"的区域
+            let contentArea = null;
+            document.querySelectorAll("h1, h2, h3, h4, span, div").forEach(el => {
+                const t = (el.textContent || "").trim();
+                if (t === "我的用户设置" || t === "My User Settings") {
+                    contentArea = el.closest("div");
+                }
+            });
+            if (!contentArea) return;
+
+            // 隐藏"登录/注册"按钮和"登录您的账户"文字
+            contentArea.querySelectorAll("button, a, span, p, div").forEach(el => {
+                const t = (el.textContent || "").trim();
+                if (t.includes("登录") || t.includes("注册") || t === "Sign In" || t === "Sign Up" || t.includes("登录您的账户") || t.includes("Log in")) {
+                    if (el.tagName === "BUTTON" || el.tagName === "A" || t.includes("登录您的账户")) {
+                        el.style.display = "none";
+                    }
+                }
+            });
+
+            // 检查是否已注入（DOM 中已存在则跳过）
+            if (document.querySelector("#o1key-apikey-box")) return;
+
+            // 创建 API Key 输入区域
+            const box = document.createElement("div");
+            box.id = "o1key-apikey-box";
+            box.style.cssText = "margin-top:24px;padding:20px;border:1px solid #444;border-radius:8px;background:#1e1e1e;";
+            box.innerHTML = `
+                <div style="font-weight:bold;font-size:15px;margin-bottom:6px;color:#eee;">O1Key API 密钥</div>
+                <div style="font-size:12px;color:#999;margin-bottom:14px;">输入您的 API 密钥，测试通过后方可保存</div>
+                <div style="display:flex;gap:8px;align-items:center;">
+                    <input id="o1key-apikey-input" type="text" placeholder="请输入 API 密钥"
+                        autocomplete="off" autocorrect="off" autocapitalize="off" spellcheck="false"
+                        data-form-type="other" data-lpignore="true" name="o1key-key-field"
+                        style="flex:1;padding:8px 12px;border:1px solid #555;border-radius:4px;background:#111;color:#eee;font-size:13px;" />
+                    <button id="o1key-apikey-test"
+                        style="padding:8px 14px;border:none;border-radius:4px;background:#47a;color:#fff;cursor:pointer;font-size:13px;white-space:nowrap;">测试令牌</button>
+                    <button id="o1key-apikey-save" disabled
+                        style="padding:8px 14px;border:none;border-radius:4px;background:#555;color:#999;cursor:not-allowed;font-size:13px;white-space:nowrap;">保存</button>
+                    <button id="o1key-apikey-clear"
+                        style="padding:8px 14px;border:none;border-radius:4px;background:#a44;color:#fff;cursor:pointer;font-size:13px;white-space:nowrap;">清空密钥</button>
+                </div>
+                <div id="o1key-apikey-status" style="margin-top:10px;font-size:12px;color:#999;"></div>
+            `;
+            contentArea.appendChild(box);
+
+            // 加载当前状态
+            try {
+                const resp = await fetch("/o1key/api_key");
+                const data = await resp.json();
+                const status = box.querySelector("#o1key-apikey-status");
+                if (data.has_key) {
+                    status.textContent = "当前密钥: " + data.masked;
+                    status.style.color = "#3b8";
+                } else {
+                    status.textContent = "尚未配置 API 密钥";
+                    status.style.color = "#a84";
+                }
+            } catch (e) {}
+
+            const saveBtn = box.querySelector("#o1key-apikey-save");
+            const testBtn = box.querySelector("#o1key-apikey-test");
+            let testPassed = false;
+
+            // 输入变化时重置测试状态
+            box.querySelector("#o1key-apikey-input").addEventListener("input", () => {
+                testPassed = false;
+                saveBtn.disabled = true;
+                saveBtn.style.background = "#555";
+                saveBtn.style.color = "#999";
+                saveBtn.style.cursor = "not-allowed";
+            });
+
+            // 测试令牌按钮
+            testBtn.addEventListener("click", async () => {
+                const input = box.querySelector("#o1key-apikey-input");
+                const status = box.querySelector("#o1key-apikey-status");
+                const key = input.value.trim();
+                if (!key) { status.textContent = "请输入密钥"; status.style.color = "#a44"; return; }
+                testBtn.disabled = true;
+                testBtn.textContent = "验证中...";
+                status.textContent = "正在验证密钥...";
+                status.style.color = "#999";
+                try {
+                    const resp = await fetch("/o1key/test_key", {
+                        method: "POST",
+                        headers: {"Content-Type": "application/json"},
+                        body: JSON.stringify({api_key: key})
+                    });
+                    const data = await resp.json();
+                    if (data.valid) {
+                        testPassed = true;
+                        status.textContent = "验证通过，可以保存";
+                        status.style.color = "#3b8";
+                        saveBtn.disabled = false;
+                        saveBtn.style.background = "#3b8";
+                        saveBtn.style.color = "#fff";
+                        saveBtn.style.cursor = "pointer";
+                    } else {
+                        testPassed = false;
+                        status.textContent = data.error || "验证失败";
+                        status.style.color = "#a44";
+                    }
+                } catch (e) {
+                    status.textContent = "网络错误";
+                    status.style.color = "#a44";
+                }
+                testBtn.disabled = false;
+                testBtn.textContent = "测试令牌";
+            });
+
+            // 保存按钮（仅测试通过后可用）
+            saveBtn.addEventListener("click", async () => {
+                if (!testPassed) return;
+                const input = box.querySelector("#o1key-apikey-input");
+                const status = box.querySelector("#o1key-apikey-status");
+                const key = input.value.trim();
+                try {
+                    const resp = await fetch("/o1key/api_key", {
+                        method: "POST",
+                        headers: {"Content-Type": "application/json"},
+                        body: JSON.stringify({api_key: key})
+                    });
+                    const data = await resp.json();
+                    if (data.success) {
+                        status.textContent = "密钥已保存";
+                        status.style.color = "#3b8";
+                        input.value = "";
+                        testPassed = false;
+                        saveBtn.disabled = true;
+                        saveBtn.style.background = "#555";
+                        saveBtn.style.color = "#999";
+                        saveBtn.style.cursor = "not-allowed";
+                    } else {
+                        status.textContent = data.error || "保存失败";
+                        status.style.color = "#a44";
+                    }
+                } catch (e) {
+                    status.textContent = "网络错误";
+                    status.style.color = "#a44";
+                }
+            });
+
+            // 清空密钥按钮
+            const clearBtn = box.querySelector("#o1key-apikey-clear");
+            clearBtn.addEventListener("click", async () => {
+                if (!confirm("确定要清空 API 密钥吗？")) return;
+                const status = box.querySelector("#o1key-apikey-status");
+                try {
+                    const resp = await fetch("/o1key/api_key", { method: "DELETE" });
+                    const data = await resp.json();
+                    if (data.success) {
+                        status.textContent = "API 密钥已清空";
+                        status.style.color = "#a84";
+                        box.querySelector("#o1key-apikey-input").value = "";
+                        testPassed = false;
+                        saveBtn.disabled = true;
+                        saveBtn.style.background = "#555";
+                        saveBtn.style.color = "#999";
+                        saveBtn.style.cursor = "not-allowed";
+                    } else {
+                        status.textContent = data.error || "清空失败";
+                        status.style.color = "#a44";
+                    }
+                } catch (e) {
+                    status.textContent = "网络错误";
+                    status.style.color = "#a44";
+                }
+            });
+        }
+
+        const observer = new MutationObserver(hide);
+        observer.observe(document.body, { childList: true, subtree: true });
+        setTimeout(hide, 1000);
+        setTimeout(hide, 3000);
+    },
+});
@@ -0,0 +1,767 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+const STORAGE_KEY = "o1key-notes";
+const SEEDED_KEY = "o1key-notes-seeded-v2";
+const NOTES_API = "/o1key/notes";
+const STYLE_ID = "o1key-notes-styles";
+
+let notes = [];
+let searchText = "";
+let activeFilter = "all";
+let noteContainer = null;
+let editingNoteId = null;
+let draftNote = null;
+let pendingDelete = false;
+
+const SAMPLE_NOTES = [
+    {
+        title: "产品主图：高级玻璃质感",
+        tags: ["产品图", "玻璃", "灯光"],
+        content: `Clean studio lighting, translucent glass material, subtle caustics, soft shadow, premium product photography, 85mm lens, minimal background, high detail.
+
+使用方式：
+1. 把产品图作为参考图输入
+2. 保留主体轮廓，只调整材质和灯光
+3. 如果玻璃过亮，降低 "caustics" 权重`
+    },
+    {
+        title: "Nano Banana 参考图经验",
+        tags: ["Nano Banana", "参考图"],
+        content: `参考图越多越容易跑偏，主体一致性优先用 1-3 张图。
+
+复杂场景建议分两步：
+1. 先生成主体和构图
+2. 再用局部或参考图做材质、背景、文字等精修
+
+如果提示词和参考图冲突，模型通常会优先参考图。`
+    },
+    {
+        title: "电商模特换装模板",
+        tags: ["电商", "模特", "换装"],
+        content: `Keep face identity and original pose. Replace the outfit with: [服装描述].
+Realistic fabric texture, natural folds, accurate seams, studio e-commerce photography, clean background, consistent lighting.
+
+Avoid changing body shape, face, hairstyle, camera angle, or hand position.`
+    },
+    {
+        title: "常用负向词",
+        tags: ["负向词", "通用"],
+        content: "low quality, blurry, deformed hands, extra fingers, bad anatomy, distorted text, watermark, logo, oversaturated, plastic skin, broken geometry"
+    },
+    {
+        title: "图片批量命名经验",
+        tags: ["批量", "工作流"],
+        content: `批量生图前先确认保存路径和命名规则。
+
+推荐流程：
+1. 小批量跑 2-3 张确认风格
+2. 固定提示词和参考图
+3. 再放大批量数量
+
+这样失败成本最低，也更容易定位是哪一环导致跑偏。`
+    }
+];
+
+const CSS = `
+#o1key-notes-root{position:absolute;inset:0;display:flex;flex-direction:column;min-height:0;color:#ddd;background:var(--comfy-menu-bg,#202020);overflow:hidden;font-family:inherit;--o1n-blue:#4f8cff;--o1n-blue-2:#7eb8f7;--o1n-blue-soft:rgba(79,140,255,.14)}
+#o1key-notes-header{padding:12px 14px 10px;border-bottom:1px solid rgba(255,255,255,.06);flex-shrink:0}
+.o1n-title-row{display:flex;align-items:center;justify-content:space-between;margin-bottom:10px}
+.o1n-title{font-size:14px;font-weight:700;color:#eee;letter-spacing:.2px}
+.o1n-head-actions{display:flex;gap:5px}
+.o1n-icon-btn{width:28px;height:28px;border:1px solid rgba(255,255,255,.09);border-radius:7px;background:rgba(255,255,255,.035);color:#888;display:flex;align-items:center;justify-content:center;cursor:pointer;transition:all .15s;flex-shrink:0}
+.o1n-icon-btn:hover{color:#ddd;background:rgba(255,255,255,.08);border-color:rgba(255,255,255,.16)}
+.o1n-icon-btn.primary{background:var(--o1n-blue);color:#fff;border-color:transparent}
+.o1n-icon-btn.primary:hover{background:#6ca0ff;color:#fff}
+#o1n-search-box{height:34px;border:1px solid rgba(255,255,255,.1);border-radius:8px;background:rgba(255,255,255,.045);display:flex;align-items:center;gap:8px;padding:0 10px;transition:border-color .15s}
+#o1n-search-box:focus-within{border-color:rgba(79,140,255,.42)}
+#o1n-search{flex:1;background:transparent;border:0;outline:0;color:#ddd;font-size:12px;min-width:0}
+#o1n-search::placeholder{color:#666}
+#o1n-panel-status{height:16px;margin-top:7px;color:#777;font-size:11px;line-height:16px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
+.o1n-tag-strip{display:flex;gap:6px;margin-top:10px;overflow-x:auto;padding-bottom:1px}
+.o1n-tag-strip::-webkit-scrollbar{display:none}
+.o1n-tag-filter{height:26px;padding:0 9px;border-radius:7px;border:1px solid rgba(255,255,255,.09);background:transparent;color:#929292;font-size:12px;display:flex;align-items:center;gap:5px;cursor:pointer;white-space:nowrap;transition:all .15s}
+.o1n-tag-filter:hover{color:#cfcfcf;border-color:rgba(255,255,255,.16)}
+.o1n-tag-filter.active{color:var(--o1n-blue-2);border-color:rgba(79,140,255,.36);background:var(--o1n-blue-soft)}
+#o1key-notes-list{flex:1;min-height:0;overflow:auto;padding:8px 10px 12px;display:flex;flex-direction:column;gap:6px}
+#o1key-notes-list::-webkit-scrollbar,#o1n-content::-webkit-scrollbar{width:4px}
+#o1key-notes-list::-webkit-scrollbar-thumb,#o1n-content::-webkit-scrollbar-thumb{background:rgba(255,255,255,.1);border-radius:2px}
+.o1n-item{border:1px solid transparent;border-radius:8px;padding:10px;background:transparent;cursor:pointer;transition:all .12s}
+.o1n-item:hover{background:rgba(255,255,255,.04);border-color:rgba(255,255,255,.07)}
+.o1n-item.active{background:rgba(255,255,255,.065);border-color:rgba(255,255,255,.12);box-shadow:inset 2px 0 0 var(--o1n-blue)}
+.o1n-item-top{display:flex;align-items:center;gap:8px;margin-bottom:6px}
+.o1n-item-title{font-size:13px;color:#e0e0e0;white-space:nowrap;overflow:hidden;text-overflow:ellipsis;flex:1;min-width:0}
+.o1n-item-edit{font-size:11px;color:#777;flex-shrink:0}
+.o1n-item-text{font-size:12px;line-height:1.45;color:#8b8b8b;display:-webkit-box;-webkit-line-clamp:3;-webkit-box-orient:vertical;overflow:hidden;word-break:break-word}
+.o1n-item-meta{margin-top:8px;display:flex;align-items:center;justify-content:space-between;gap:8px;color:#666;font-size:10px}
+.o1n-item-tags{display:flex;gap:4px;overflow:hidden;min-width:0}
+.o1n-tag{color:var(--o1n-blue-2);background:rgba(79,140,255,.1);border:1px solid rgba(79,140,255,.2);border-radius:5px;padding:2px 5px;white-space:nowrap;max-width:120px;overflow:hidden;text-overflow:ellipsis}
+.o1n-empty{height:100%;display:flex;align-items:center;justify-content:center;text-align:center;color:#666;font-size:13px;line-height:1.6;padding:20px}
+.o1n-editor-backdrop{position:absolute;inset:0;background:rgba(0,0,0,.34);display:flex;align-items:stretch;justify-content:flex-end;z-index:30;animation:o1n-fade .12s ease}
+.o1n-editor{width:100%;height:100%;background:var(--comfy-menu-bg,#202020);border-left:1px solid rgba(79,140,255,.28);box-shadow:0 22px 70px rgba(0,0,0,.46);display:flex;flex-direction:column;min-height:0}
+.o1n-editor-top{height:48px;padding:0 14px;border-bottom:1px solid rgba(255,255,255,.08);display:flex;align-items:center;justify-content:space-between;flex-shrink:0}
+.o1n-editor-title{font-size:13px;font-weight:700;color:#eee}
+.o1n-editor-body{padding:12px;display:flex;flex-direction:column;gap:9px;flex:1;min-height:0}
+#o1n-title-input,#o1n-content{width:100%;border:1px solid rgba(255,255,255,.1);border-radius:8px;background:rgba(255,255,255,.045);color:#ddd;outline:0;font-family:inherit;transition:border-color .15s}
+#o1n-title-input{height:34px;padding:0 10px;font-size:13px;font-weight:700;flex-shrink:0}
+#o1n-content{flex:1;min-height:160px;resize:none;padding:11px 12px;font-size:13px;line-height:1.58}
+#o1n-title-input:focus,#o1n-content:focus{border-color:rgba(79,140,255,.42)}
+.o1n-tag-editor{min-height:74px;border:1px solid rgba(255,255,255,.1);border-radius:8px;background:rgba(255,255,255,.035);padding:7px;flex-shrink:0}
+.o1n-tag-row{display:flex;align-items:center;gap:6px;flex-wrap:wrap}
+.o1n-tag-token{height:24px;padding:0 7px;border-radius:6px;border:1px solid rgba(79,140,255,.25);background:rgba(79,140,255,.1);color:var(--o1n-blue-2);display:inline-flex;align-items:center;gap:6px;font-size:12px}
+.o1n-tag-remove{color:#8baee8;font-size:13px;line-height:1;cursor:pointer}
+.o1n-tag-input{height:24px;min-width:104px;flex:1;border:1px dashed rgba(255,255,255,.14);border-radius:6px;background:transparent;color:#ddd;padding:0 7px;font-size:12px;outline:0}
+.o1n-tag-input::placeholder{color:#777}
+.o1n-tag-hint{margin-top:7px;color:#666;font-size:11px;line-height:1.35}
+.o1n-editor-actions{display:grid;grid-template-columns:1fr 1fr 1fr;gap:7px;padding:12px;border-top:1px solid rgba(255,255,255,.08);flex-shrink:0}
+.o1n-action{height:32px;border:1px solid rgba(255,255,255,.1);border-radius:7px;background:rgba(255,255,255,.035);color:#aaa;font-size:12px;display:flex;align-items:center;justify-content:center;gap:6px;cursor:pointer;transition:all .15s}
+.o1n-action:hover{color:#e5e5e5;background:rgba(255,255,255,.075);border-color:rgba(255,255,255,.16)}
+.o1n-action.primary{background:var(--o1n-blue);color:#fff;border-color:transparent;font-weight:700}
+.o1n-action.primary:hover{background:#6ca0ff;color:#fff}
+.o1n-action.accent{background:#d8b45b;color:#171717;border-color:transparent;font-weight:700}
+.o1n-action.accent:hover{background:#e3c474;color:#111}
+.o1n-action.danger:hover{background:rgba(215,101,101,.18);border-color:rgba(215,101,101,.32);color:#f0a0a0}
+.o1n-delete-confirm{grid-column:1 / -1;display:none;align-items:center;gap:7px;padding:7px 8px;border:1px solid rgba(215,101,101,.24);border-radius:7px;background:rgba(215,101,101,.08);color:#ccc;font-size:12px}
+.o1n-delete-confirm.show{display:flex}
+.o1n-delete-confirm span{flex:1;min-width:0}
+.o1n-mini-btn{height:24px;border:1px solid rgba(255,255,255,.1);border-radius:6px;background:rgba(255,255,255,.06);color:#bbb;padding:0 8px;font-size:11px;cursor:pointer}
+.o1n-mini-btn.danger{background:rgba(215,101,101,.5);border-color:rgba(215,101,101,.3);color:#fff}
+#o1n-status{grid-column:1 / -1;height:16px;color:#777;font-size:11px;line-height:16px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
+#o1n-import-file{display:none}
+@keyframes o1n-fade{from{opacity:0}to{opacity:1}}
+`;
+
+function genId() {
+    return Date.now().toString(36) + Math.random().toString(36).slice(2, 8);
+}
+
+function now() {
+    return Date.now();
+}
+
+function parseTags(value) {
+    if (Array.isArray(value)) return value.map(String).map(s => s.trim()).filter(Boolean);
+    return String(value || "")
+        .split(/[,，、\s]+/)
+        .map(s => s.trim())
+        .filter(Boolean);
+}
+
+function makeNote(data = {}) {
+    const ts = now();
+    return {
+        id: data.id || genId(),
+        title: String(data.title || "未命名笔记"),
+        tags: parseTags(data.tags || data.category || ""),
+        content: String(data.content || ""),
+        createdAt: data.createdAt || ts,
+        updatedAt: data.updatedAt || ts,
+    };
+}
+
+function cloneNote(note) {
+    return {
+        ...note,
+        tags: [...(note.tags || [])],
+    };
+}
+
+function formatDate(ts) {
+    const d = new Date(ts || now());
+    const today = new Date();
+    const pad = n => String(n).padStart(2, "0");
+    if (d.toDateString() === today.toDateString()) {
+        return `今天 ${pad(d.getHours())}:${pad(d.getMinutes())}`;
+    }
+    return `${pad(d.getMonth() + 1)}/${pad(d.getDate())}`;
+}
+
+function escapeHtml(value) {
+    return String(value ?? "")
+        .replace(/&/g, "&amp;")
+        .replace(/</g, "&lt;")
+        .replace(/>/g, "&gt;")
+        .replace(/"/g, "&quot;")
+        .replace(/'/g, "&#039;");
+}
+
+function icon(name, size = 14) {
+    const icons = {
+        plus: `<path d="M12 5v14M5 12h14"/>`,
+        search: `<circle cx="11" cy="11" r="7"/><path d="M20 20l-3.5-3.5"/>`,
+        copy: `<rect x="9" y="9" width="13" height="13" rx="2"/><path d="M5 15H4a2 2 0 01-2-2V4a2 2 0 012-2h9a2 2 0 012 2v1"/>`,
+        insert: `<path d="M12 5v14"/><path d="M19 12H5"/>`,
+        save: `<path d="M19 21H5a2 2 0 01-2-2V5a2 2 0 012-2h11l5 5v11a2 2 0 01-2 2z"/><path d="M17 21v-8H7v8"/><path d="M7 3v5h8"/>`,
+        trash: `<path d="M3 6h18"/><path d="M8 6V4h8v2"/><path d="M19 6l-1 14H6L5 6"/>`,
+        download: `<path d="M21 15v4a2 2 0 01-2 2H5a2 2 0 01-2-2v-4"/><path d="M7 10l5 5 5-5"/><path d="M12 15V3"/>`,
+        upload: `<path d="M21 15v4a2 2 0 01-2 2H5a2 2 0 01-2-2v-4"/><path d="M17 8l-5-5-5 5"/><path d="M12 3v12"/>`,
+        node: `<rect x="3" y="4" width="7" height="7" rx="1"/><rect x="14" y="13" width="7" height="7" rx="1"/><path d="M10 7.5h3a4 4 0 014 4V13"/>`,
+        close: `<path d="M18 6L6 18M6 6l12 12"/>`,
+    };
+    return `<svg width="${size}" height="${size}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">${icons[name] || icons.plus}</svg>`;
+}
+
+function injectStyles() {
+    if (document.getElementById(STYLE_ID)) return;
+    const el = document.createElement("style");
+    el.id = STYLE_ID;
+    el.textContent = CSS;
+    document.head.appendChild(el);
+}
+
+function readCachedNotes() {
+    try {
+        const raw = localStorage.getItem(STORAGE_KEY);
+        if (raw === null) return { found: false, notes: [] };
+        const parsed = JSON.parse(raw);
+        return { found: true, notes: Array.isArray(parsed) ? parsed.map(makeNote) : [] };
+    } catch {
+        return { found: false, notes: [] };
+    }
+}
+
+function writeCachedNotes() {
+    try {
+        localStorage.setItem(STORAGE_KEY, JSON.stringify(notes));
+    } catch {}
+}
+
+function hasSeededNotes() {
+    try {
+        return !!localStorage.getItem(SEEDED_KEY);
+    } catch {
+        return false;
+    }
+}
+
+function markSeededNotes() {
+    try {
+        localStorage.setItem(SEEDED_KEY, "1");
+    } catch {}
+}
+
+function createSeedNotes() {
+    return SAMPLE_NOTES.map(makeNote);
+}
+
+async function loadNotes() {
+    const cached = readCachedNotes();
+
+    try {
+        const resp = await api.fetchApi(NOTES_API);
+        if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
+        const data = await resp.json();
+
+        if (data.exists) {
+            notes = Array.isArray(data.notes) ? data.notes.map(makeNote) : [];
+            writeCachedNotes();
+            return;
+        }
+
+        notes = cached.found ? cached.notes : createSeedNotes();
+        markSeededNotes();
+        writeCachedNotes();
+        await persistNotesToFile();
+        return;
+    } catch (e) {
+        console.warn("[o1key notes] file storage unavailable, using localStorage", e);
+    }
+
+    notes = cached.found ? cached.notes : [];
+    if (!cached.found && !hasSeededNotes()) {
+        notes = createSeedNotes();
+        markSeededNotes();
+        writeCachedNotes();
+    }
+}
+
+function saveNotes() {
+    writeCachedNotes();
+    void persistNotesToFile();
+}
+
+async function persistNotesToFile() {
+    try {
+        const resp = await api.fetchApi(NOTES_API, {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({ notes }),
+        });
+        if (!resp.ok) throw new Error(`HTTP ${resp.status}`);
+        return true;
+    } catch (e) {
+        console.warn("[o1key notes] failed to save notes file", e);
+        setPanelStatus("笔记文件保存失败，已保存在浏览器缓存");
+        return false;
+    }
+}
+
+function allTags() {
+    const counts = new Map();
+    for (const note of notes) {
+        for (const tag of note.tags || []) counts.set(tag, (counts.get(tag) || 0) + 1);
+    }
+    return [...counts.entries()].sort((a, b) => b[1] - a[1]).slice(0, 12);
+}
+
+function filteredNotes() {
+    const q = searchText.trim().toLowerCase();
+    return notes.filter(note => {
+        if (activeFilter.startsWith("tag:") && !(note.tags || []).includes(activeFilter.slice(4))) return false;
+        if (!q) return true;
+        return [note.title, note.content, ...(note.tags || [])].join("\n").toLowerCase().includes(q);
+    }).sort((a, b) => (b.updatedAt || 0) - (a.updatedAt || 0));
+}
+
+app.registerExtension({
+    name: "o1key.notePanel",
+    async setup() {
+        await loadNotes();
+        app.extensionManager.registerSidebarTab({
+            id: "o1key-notes",
+            title: "笔记",
+            icon: "pi pi-pencil",
+            type: "custom",
+            render: (container) => {
+                injectStyles();
+                renderNotePanel(container);
+            },
+        });
+    },
+});
+
+function renderNotePanel(container) {
+    container.innerHTML = "";
+    container.style.position = "relative";
+    container.style.height = "100%";
+    container.style.overflow = "hidden";
+
+    const root = document.createElement("div");
+    root.id = "o1key-notes-root";
+    root.innerHTML = `
+        <div id="o1key-notes-header">
+            <div class="o1n-title-row">
+                <div class="o1n-title">笔记</div>
+                <div class="o1n-head-actions">
+                    <button class="o1n-icon-btn" id="o1n-import" title="导入 JSON">${icon("upload")}</button>
+                    <button class="o1n-icon-btn" id="o1n-export" title="导出 JSON">${icon("download")}</button>
+                    <button class="o1n-icon-btn primary" id="o1n-new" title="新建笔记">${icon("plus")}</button>
+                </div>
+            </div>
+            <div id="o1n-search-box">${icon("search", 13)}<input id="o1n-search" placeholder="搜索笔记内容或标签" value="${escapeHtml(searchText)}"></div>
+            <div class="o1n-tag-strip" id="o1n-tag-strip"></div>
+            <div id="o1n-panel-status"></div>
+        </div>
+        <div id="o1key-notes-list"></div>
+        <input id="o1n-import-file" type="file" accept=".json,application/json">
+    `;
+    container.appendChild(root);
+    noteContainer = root;
+    bindPanelEvents(root);
+    renderTagFilters();
+    renderList();
+    renderEditor();
+}
+
+function bindPanelEvents(root) {
+    root.querySelector("#o1n-new").addEventListener("click", createNote);
+    root.querySelector("#o1n-export").addEventListener("click", exportNotes);
+    root.querySelector("#o1n-import").addEventListener("click", () => root.querySelector("#o1n-import-file").click());
+    root.querySelector("#o1n-import-file").addEventListener("change", importNotes);
+    root.querySelector("#o1n-search").addEventListener("input", (e) => {
+        searchText = e.target.value;
+        renderList();
+    });
+}
+
+function renderTagFilters() {
+    const row = noteContainer?.querySelector("#o1n-tag-strip");
+    if (!row) return;
+    const tags = allTags();
+    row.innerHTML = `<button class="o1n-tag-filter${activeFilter === "all" ? " active" : ""}" data-filter="all">全部 <span>${notes.length}</span></button>` +
+        tags.map(([tag, count]) => {
+            const filter = `tag:${tag}`;
+            return `<button class="o1n-tag-filter${activeFilter === filter ? " active" : ""}" data-filter="${escapeHtml(filter)}">#${escapeHtml(tag)} <span>${count}</span></button>`;
+        }).join("");
+
+    row.querySelectorAll("[data-filter]").forEach(btn => {
+        btn.addEventListener("click", () => {
+            activeFilter = btn.dataset.filter;
+            renderTagFilters();
+            renderList();
+        });
+    });
+}
+
+function renderList() {
+    const list = noteContainer.querySelector("#o1key-notes-list");
+    const visible = filteredNotes();
+
+    if (!visible.length) {
+        list.innerHTML = `<div class="o1n-empty">没有匹配的笔记</div>`;
+        return;
+    }
+
+    list.innerHTML = visible.map(note => {
+        const tags = (note.tags || []).slice(0, 3).map(tag => `<span class="o1n-tag">#${escapeHtml(tag)}</span>`).join("");
+        return `<div class="o1n-item${note.id === editingNoteId ? " active" : ""}" data-id="${note.id}">
+            <div class="o1n-item-top">
+                <div class="o1n-item-title">${escapeHtml(note.title || "未命名笔记")}</div>
+                <div class="o1n-item-edit">编辑</div>
+            </div>
+            <div class="o1n-item-text">${escapeHtml(note.content || "空笔记")}</div>
+            <div class="o1n-item-meta"><div class="o1n-item-tags">${tags || `<span class="o1n-tag">#未分类</span>`}</div><span>${formatDate(note.updatedAt)}</span></div>
+        </div>`;
+    }).join("");
+
+    list.querySelectorAll(".o1n-item").forEach(item => {
+        item.addEventListener("click", () => openEditor(item.dataset.id));
+    });
+}
+
+function renderEditor() {
+    noteContainer.querySelector(".o1n-editor-backdrop")?.remove();
+    if (!draftNote) return;
+    pendingDelete = false;
+
+    const overlay = document.createElement("div");
+    overlay.className = "o1n-editor-backdrop";
+    overlay.innerHTML = `
+        <div class="o1n-editor">
+            <div class="o1n-editor-top">
+                <div class="o1n-editor-title">${editingNoteId ? "编辑笔记" : "新建笔记"}</div>
+                <button class="o1n-icon-btn" id="o1n-close-editor" title="关闭">${icon("close", 13)}</button>
+            </div>
+            <div class="o1n-editor-body">
+                <input id="o1n-title-input" value="${escapeHtml(draftNote.title)}" placeholder="笔记标题">
+                <div class="o1n-tag-editor">
+                    <div class="o1n-tag-row">
+                        ${(draftNote.tags || []).map(tag => `<span class="o1n-tag-token">${escapeHtml(tag)} <span class="o1n-tag-remove" data-tag="${escapeHtml(tag)}">×</span></span>`).join("")}
+                        <input class="o1n-tag-input" id="o1n-tag-input" placeholder="＋ 添加标签">
+                    </div>
+                    <div class="o1n-tag-hint">输入标签后按 Enter，或用逗号分隔多个标签。</div>
+                </div>
+                <textarea id="o1n-content" placeholder="记录提示词、参数经验、踩坑结论...">${escapeHtml(draftNote.content)}</textarea>
+            </div>
+            <div class="o1n-editor-actions">
+                <button class="o1n-action" id="o1n-cancel">取消</button>
+                <button class="o1n-action" id="o1n-copy">${icon("copy", 13)}复制</button>
+                <button class="o1n-action accent" id="o1n-insert">${icon("insert", 13)}插入</button>
+                <button class="o1n-action" id="o1n-save-from-node">${icon("node", 13)}从节点保存</button>
+                <button class="o1n-action danger" id="o1n-delete">${icon("trash", 13)}删除</button>
+                <button class="o1n-action primary" id="o1n-save">${icon("save", 13)}保存</button>
+                <div class="o1n-delete-confirm" id="o1n-delete-confirm">
+                    <span>确定删除这条笔记？</span>
+                    <button class="o1n-mini-btn danger" id="o1n-delete-yes">删除</button>
+                    <button class="o1n-mini-btn" id="o1n-delete-no">取消</button>
+                </div>
+                <div id="o1n-status"></div>
+            </div>
+        </div>
+    `;
+    noteContainer.appendChild(overlay);
+    bindEditorEvents(overlay);
+}
+
+function bindEditorEvents(overlay) {
+    overlay.querySelector("#o1n-close-editor").addEventListener("click", closeEditor);
+    overlay.querySelector("#o1n-cancel").addEventListener("click", closeEditor);
+    overlay.querySelector("#o1n-title-input").addEventListener("input", updateDraftFromEditor);
+    overlay.querySelector("#o1n-content").addEventListener("input", updateDraftFromEditor);
+    overlay.querySelectorAll(".o1n-tag-remove").forEach(btn => {
+        btn.addEventListener("click", () => removeDraftTag(btn.dataset.tag));
+    });
+    overlay.querySelector("#o1n-tag-input").addEventListener("keydown", handleTagInputKeydown);
+    overlay.querySelector("#o1n-tag-input").addEventListener("blur", commitTagInput);
+    overlay.querySelector("#o1n-copy").addEventListener("click", copyDraftContent);
+    overlay.querySelector("#o1n-insert").addEventListener("click", insertDraftContent);
+    overlay.querySelector("#o1n-save-from-node").addEventListener("click", saveFromCurrentNode);
+    overlay.querySelector("#o1n-delete").addEventListener("click", requestDeleteEditingNote);
+    overlay.querySelector("#o1n-delete-yes").addEventListener("click", deleteEditingNote);
+    overlay.querySelector("#o1n-delete-no").addEventListener("click", hideDeleteConfirm);
+    overlay.querySelector("#o1n-save").addEventListener("click", saveDraft);
+}
+
+function openEditor(id) {
+    const note = notes.find(n => n.id === id);
+    if (!note) return;
+    editingNoteId = id;
+    draftNote = cloneNote(note);
+    renderList();
+    renderEditor();
+}
+
+function closeEditor() {
+    editingNoteId = null;
+    draftNote = null;
+    pendingDelete = false;
+    renderList();
+    renderEditor();
+}
+
+function createNote() {
+    editingNoteId = null;
+    draftNote = makeNote({ title: "新笔记", tags: ["未分类"], content: "" });
+    renderList();
+    renderEditor();
+    noteContainer.querySelector("#o1n-title-input")?.focus();
+}
+
+function updateDraftFromEditor() {
+    if (!draftNote) return;
+    draftNote.title = noteContainer.querySelector("#o1n-title-input")?.value.trim() || "未命名笔记";
+    draftNote.content = noteContainer.querySelector("#o1n-content")?.value || "";
+    draftNote.updatedAt = now();
+}
+
+function addTagsFromText(value) {
+    if (!draftNote) return false;
+    const incoming = parseTags(value);
+    if (!incoming.length) return false;
+    const existing = new Set(draftNote.tags || []);
+    for (const tag of incoming) existing.add(tag);
+    draftNote.tags = [...existing];
+    draftNote.updatedAt = now();
+    renderDraftTags();
+    return true;
+}
+
+function commitTagInput() {
+    const input = noteContainer?.querySelector("#o1n-tag-input");
+    if (!input) return;
+    if (addTagsFromText(input.value)) input.value = "";
+}
+
+function handleTagInputKeydown(e) {
+    if (e.key !== "Enter" && e.key !== "," && e.key !== "，") return;
+    e.preventDefault();
+    commitTagInput();
+}
+
+function removeDraftTag(tag) {
+    if (!draftNote) return;
+    draftNote.tags = (draftNote.tags || []).filter(t => t !== tag);
+    draftNote.updatedAt = now();
+    renderDraftTags();
+}
+
+function renderDraftTags() {
+    const row = noteContainer?.querySelector(".o1n-tag-row");
+    if (!row || !draftNote) return;
+    row.innerHTML = `
+        ${(draftNote.tags || []).map(tag => `<span class="o1n-tag-token">${escapeHtml(tag)} <span class="o1n-tag-remove" data-tag="${escapeHtml(tag)}">×</span></span>`).join("")}
+        <input class="o1n-tag-input" id="o1n-tag-input" placeholder="＋ 添加标签">
+    `;
+    row.querySelectorAll(".o1n-tag-remove").forEach(btn => {
+        btn.addEventListener("click", () => removeDraftTag(btn.dataset.tag));
+    });
+    const input = row.querySelector("#o1n-tag-input");
+    input.addEventListener("keydown", handleTagInputKeydown);
+    input.addEventListener("blur", commitTagInput);
+    input.focus();
+}
+
+function saveDraft() {
+    if (!draftNote) return;
+    updateDraftFromEditor();
+    draftNote.tags = draftNote.tags?.length ? draftNote.tags : ["未分类"];
+    if (editingNoteId) {
+        const idx = notes.findIndex(n => n.id === editingNoteId);
+        if (idx >= 0) notes[idx] = cloneNote(draftNote);
+    } else {
+        draftNote.id = genId();
+        draftNote.createdAt = now();
+        draftNote.updatedAt = now();
+        notes.unshift(cloneNote(draftNote));
+        editingNoteId = draftNote.id;
+    }
+    saveNotes();
+    renderTagFilters();
+    closeEditor();
+}
+
+function requestDeleteEditingNote() {
+    if (!editingNoteId) {
+        closeEditor();
+        return;
+    }
+    pendingDelete = true;
+    const box = noteContainer?.querySelector("#o1n-delete-confirm");
+    box?.classList.add("show");
+    setStatus("");
+}
+
+function hideDeleteConfirm() {
+    pendingDelete = false;
+    noteContainer?.querySelector("#o1n-delete-confirm")?.classList.remove("show");
+}
+
+function deleteEditingNote() {
+    if (!editingNoteId) {
+        closeEditor();
+        return;
+    }
+    const note = notes.find(n => n.id === editingNoteId);
+    if (!note) return;
+    notes = notes.filter(n => n.id !== editingNoteId);
+    saveNotes();
+    renderTagFilters();
+    closeEditor();
+}
+
+async function copyDraftContent() {
+    updateDraftFromEditor();
+    try {
+        await navigator.clipboard.writeText(draftNote?.content || "");
+        setStatus("已复制");
+    } catch {
+        setStatus("复制失败，请手动选择内容");
+    }
+}
+
+function insertDraftContent() {
+    updateDraftFromEditor();
+    const text = draftNote?.content || "";
+    if (!text.trim()) {
+        setStatus("当前笔记内容为空");
+        return;
+    }
+    if (insertIntoFocusedInput(text) || insertIntoSelectedNode(text)) {
+        setStatus("已插入");
+        return;
+    }
+    navigator.clipboard?.writeText(text).catch(() => {});
+    setStatus("未找到可插入位置，已复制内容");
+}
+
+function setStatus(message) {
+    const el = noteContainer?.querySelector("#o1n-status");
+    if (!el) return;
+    el.textContent = message;
+    clearTimeout(setStatus._timer);
+    setStatus._timer = setTimeout(() => {
+        if (el.textContent === message) el.textContent = "";
+    }, 2200);
+}
+
+function setPanelStatus(message) {
+    const el = noteContainer?.querySelector("#o1n-panel-status");
+    if (!el) return;
+    el.textContent = message;
+    clearTimeout(setPanelStatus._timer);
+    setPanelStatus._timer = setTimeout(() => {
+        if (el.textContent === message) el.textContent = "";
+    }, 2600);
+}
+
+function insertIntoFocusedInput(text) {
+    const el = document.activeElement;
+    if (!el || noteContainer.contains(el)) return false;
+    if (!(el instanceof HTMLTextAreaElement || el instanceof HTMLInputElement)) return false;
+    const start = el.selectionStart ?? el.value.length;
+    const end = el.selectionEnd ?? el.value.length;
+    const before = el.value.slice(0, start);
+    const after = el.value.slice(end);
+    const spacer = before && !before.endsWith("\n") ? "\n" : "";
+    el.value = before + spacer + text + after;
+    const pos = (before + spacer + text).length;
+    el.setSelectionRange(pos, pos);
+    el.dispatchEvent(new Event("input", { bubbles: true }));
+    el.dispatchEvent(new Event("change", { bubbles: true }));
+    return true;
+}
+
+function findSelectedNode() {
+    const selectedNodes = app.canvas?.selected_nodes;
+    if (selectedNodes) {
+        const values = Array.isArray(selectedNodes) ? selectedNodes : Object.values(selectedNodes);
+        if (values.length) return values[0];
+    }
+    return app.canvas?.selected_node || null;
+}
+
+function isPromptWidget(widget) {
+    const name = String(widget?.name || "").toLowerCase();
+    const value = widget?.value;
+    if (typeof value !== "string") return false;
+    return (
+        name.includes("prompt") ||
+        name.includes("提示词") ||
+        name.includes("正向") ||
+        name === "text" ||
+        name === "文本"
+    );
+}
+
+function insertIntoSelectedNode(text) {
+    const node = findSelectedNode();
+    if (!node?.widgets?.length) return false;
+    const widget = node.widgets.find(isPromptWidget) || node.widgets.find(w => typeof w.value === "string");
+    if (!widget) return false;
+    const current = widget.value || "";
+    const spacer = current && !current.endsWith("\n") ? "\n" : "";
+    widget.value = current + spacer + text;
+    widget.callback?.(widget.value, app.canvas, node, widget);
+    app.graph?.setDirtyCanvas?.(true, true);
+    return true;
+}
+
+function getPromptWidgetsFromSelectedNode() {
+    const node = findSelectedNode();
+    if (!node?.widgets?.length) return [];
+    return node.widgets.filter(w => typeof w.value === "string" && String(w.value).trim());
+}
+
+function saveFromCurrentNode() {
+    const widgets = getPromptWidgetsFromSelectedNode();
+    if (!widgets.length) {
+        setStatus("当前没有选中包含文本的节点");
+        return;
+    }
+
+    const preferred = widgets.find(isPromptWidget) || widgets[0];
+    const node = findSelectedNode();
+    draftNote = makeNote({
+        title: `${node?.title || node?.type || "节点"}：${preferred.name || "提示词"}`,
+        tags: [node?.title || node?.type || "节点"],
+        content: preferred.value,
+    });
+    editingNoteId = null;
+    renderEditor();
+    setStatus("已读取当前节点内容，保存后写入笔记");
+}
+
+function exportNotes() {
+    const blob = new Blob([JSON.stringify(notes, null, 2)], { type: "application/json" });
+    const url = URL.createObjectURL(blob);
+    const a = document.createElement("a");
+    a.href = url;
+    a.download = `o1key-notes-${new Date().toISOString().slice(0, 10)}.json`;
+    document.body.appendChild(a);
+    a.click();
+    a.remove();
+    URL.revokeObjectURL(url);
+}
+
+async function importNotes(e) {
+    const file = e.target.files?.[0];
+    e.target.value = "";
+    if (!file) return;
+    try {
+        const imported = JSON.parse(await file.text());
+        const list = Array.isArray(imported) ? imported : imported.notes;
+        if (!Array.isArray(list)) throw new Error("Invalid notes file");
+        const existing = new Set(notes.map(n => n.id));
+        const normalized = list.map(makeNote).map(n => {
+            if (existing.has(n.id)) n.id = genId();
+            return n;
+        });
+        notes = [...normalized, ...notes];
+        saveNotes();
+        renderTagFilters();
+        renderList();
+        setPanelStatus(`已导入 ${normalized.length} 条笔记`);
+    } catch {
+        setPanelStatus("导入失败，请确认 JSON 格式");
+    }
+}
@@ -0,0 +1,797 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+// Fabric.js local loader
+let fabricLoaded = false;
+function loadFabric() {
+  if (fabricLoaded) return Promise.resolve();
+  return new Promise((resolve, reject) => {
+    const script = document.createElement("script");
+    // Load from local extension directory (allowed by CSP 'self')
+    script.src = new URL("./lib/fabric.min.js", import.meta.url).href;
+    script.onload = () => { fabricLoaded = true; resolve(); };
+    script.onerror = () => reject(new Error("Failed to load Fabric.js"));
+    document.head.appendChild(script);
+  });
+}
+
+// --- Styles ---
+const STYLES = `
+.pb-overlay { position:fixed; inset:0; z-index:99999; background:rgba(0,0,0,0.85); display:flex; flex-direction:column; align-items:center; justify-content:center; }
+.pb-toolbar { display:flex; gap:6px; padding:10px 16px; background:#1e1e1e; border-radius:8px; margin-bottom:10px; align-items:center; flex-wrap:wrap; }
+.pb-toolbar button { background:#333; color:#eee; border:1px solid #555; border-radius:4px; padding:6px 12px; cursor:pointer; font-size:13px; transition:all .15s; }
+.pb-toolbar button:hover { background:#444; }
+.pb-toolbar button.active { background:#0066ff; border-color:#0066ff; color:#fff; }
+.pb-toolbar .pb-sep { width:1px; height:24px; background:#555; margin:0 4px; }
+.pb-toolbar input[type=color] { width:32px; height:28px; border:none; padding:0; cursor:pointer; border-radius:4px; }
+.pb-toolbar input[type=range] { width:80px; accent-color:#0066ff; }
+.pb-toolbar input.pb-mosaic-size { width:90px; }
+.pb-toolbar .pb-label { color:#aaa; font-size:12px; }
+.pb-canvas-wrap { border:2px solid #444; border-radius:4px; overflow:hidden; }
+.pb-actions { display:flex; gap:10px; margin-top:10px; }
+.pb-actions button { padding:8px 24px; border-radius:6px; font-size:14px; cursor:pointer; border:none; }
+.pb-actions .pb-cancel { background:#555; color:#eee; }
+.pb-actions .pb-confirm { background:#0066ff; color:#fff; }
+`;
+
+function injectStyles() {
+  if (document.getElementById("pb-styles")) return;
+  const el = document.createElement("style");
+  el.id = "pb-styles";
+  el.textContent = STYLES;
+  document.head.appendChild(el);
+}
+
+function clamp(value, min, max) {
+  return Math.max(min, Math.min(max, value));
+}
+
+const PB_EXPORT_PROPS = ["pbTool"];
+
+function clampPointer(canvas, pointer) {
+  return {
+    x: clamp(pointer.x, 0, canvas.width),
+    y: clamp(pointer.y, 0, canvas.height),
+  };
+}
+
+function makeCircleCursor(size) {
+  const cursorSize = clamp(Math.round(size), 18, 80);
+  const center = cursorSize / 2;
+  const hotspot = Math.round(center);
+  const radius = Math.max(3, center - 2);
+  const svg = `
+    <svg xmlns="http://www.w3.org/2000/svg" width="${cursorSize}" height="${cursorSize}" viewBox="0 0 ${cursorSize} ${cursorSize}">
+      <circle cx="${center}" cy="${center}" r="${radius}" fill="none" stroke="black" stroke-width="3"/>
+      <circle cx="${center}" cy="${center}" r="${radius}" fill="none" stroke="white" stroke-width="1.5"/>
+    </svg>
+  `.trim();
+  return `url("data:image/svg+xml;charset=UTF-8,${encodeURIComponent(svg)}") ${hotspot} ${hotspot}, crosshair`;
+}
+
+function applyMosaicCursor(canvas, state) {
+  const cursor = makeCircleCursor(state.mosaicSize);
+  canvas.defaultCursor = cursor;
+  canvas.hoverCursor = cursor;
+}
+
+function configureMosaicObject(obj) {
+  obj.set({
+    selectable: false,
+    hasControls: false,
+    hasBorders: false,
+    lockMovementX: true,
+    lockMovementY: true,
+    lockScalingX: true,
+    lockScalingY: true,
+    lockRotation: true,
+    perPixelTargetFind: true,
+    objectCaching: false,
+  });
+  obj.pbTool = "mosaic";
+  return obj;
+}
+
+function normalizeMosaicObjects(canvas) {
+  canvas.getObjects().forEach((obj) => {
+    if (obj.pbTool === "mosaic" || obj.type === "image") {
+      configureMosaicObject(obj);
+    }
+  });
+}
+
+// --- Get current image URL from node ---
+function getImageUrl(node) {
+  if (node.imgs && node.imgs.length > 0) {
+    return node.imgs[node.imageIndex ?? 0].src;
+  }
+  const widget = node.widgets?.find(w => w.name === "image");
+  if (!widget?.value) return null;
+  const val = String(widget.value);
+  const match = val.match(/^(.+?)(?:\s*\[(\w+)\])?$/);
+  if (!match) return null;
+  const filename = match[1];
+  const type = match[2] || "input";
+  const parts = filename.split("/");
+  const name = parts.pop();
+  const subfolder = parts.join("/");
+  return `/view?filename=${encodeURIComponent(name)}&type=${type}&subfolder=${encodeURIComponent(subfolder)}`;
+}
+
+// --- History Manager ---
+class HistoryManager {
+  constructor(canvas) {
+    this.canvas = canvas;
+    this.stack = [];
+    this.index = -1;
+    this.locked = false;
+  }
+  save() {
+    if (this.locked) return;
+    this.index++;
+    this.stack.length = this.index;
+    this.stack.push(this.canvas.toJSON(PB_EXPORT_PROPS));
+  }
+  undo() {
+    if (this.index <= 0) return;
+    this.index--;
+    this._restore();
+  }
+  redo() {
+    if (this.index >= this.stack.length - 1) return;
+    this.index++;
+    this._restore();
+  }
+  _restore() {
+    this.locked = true;
+    this.canvas.loadFromJSON(this.stack[this.index], () => {
+      normalizeMosaicObjects(this.canvas);
+      this.canvas.renderAll();
+      this.locked = false;
+    });
+  }
+}
+
+// --- Shape drawing handler ---
+function setupShapeDrawing(canvas, state) {
+  let startX, startY, shape;
+
+  canvas.on("mouse:down", (opt) => {
+    if (state.tool === "select" || state.tool === "brush" || state.tool === "eraser" || state.tool === "mosaic") return;
+    const ptr = canvas.getPointer(opt.e);
+    startX = ptr.x;
+    startY = ptr.y;
+    state.drawing = true;
+
+    const opts = { left: startX, top: startY, fill: "transparent", stroke: state.color, strokeWidth: state.width, selectable: true };
+
+    if (state.tool === "rect") {
+      shape = new fabric.Rect({ ...opts, width: 0, height: 0 });
+    } else if (state.tool === "circle") {
+      shape = new fabric.Ellipse({ ...opts, rx: 0, ry: 0 });
+    } else if (state.tool === "line") {
+      shape = new fabric.Line([startX, startY, startX, startY], { stroke: state.color, strokeWidth: state.width, selectable: true });
+    }
+    if (shape) canvas.add(shape);
+  });
+
+  canvas.on("mouse:move", (opt) => {
+    if (!state.drawing || !shape) return;
+    const ptr = canvas.getPointer(opt.e);
+    const dx = ptr.x - startX;
+    const dy = ptr.y - startY;
+
+    if (state.tool === "rect") {
+      shape.set({ left: dx > 0 ? startX : ptr.x, top: dy > 0 ? startY : ptr.y, width: Math.abs(dx), height: Math.abs(dy) });
+    } else if (state.tool === "circle") {
+      shape.set({ left: dx > 0 ? startX : ptr.x, top: dy > 0 ? startY : ptr.y, rx: Math.abs(dx) / 2, ry: Math.abs(dy) / 2 });
+    } else if (state.tool === "line") {
+      shape.set({ x2: ptr.x, y2: ptr.y });
+    }
+    canvas.renderAll();
+  });
+
+  canvas.on("mouse:up", () => {
+    if (!state.drawing || !shape) return;
+    state.drawing = false;
+    shape = null;
+  });
+}
+
+// --- Mosaic brush handler ---
+function createMosaicSource(sourceImg, canvas, blockSize) {
+  const scaleX = canvas.width / sourceImg.width;
+  const scaleY = canvas.height / sourceImg.height;
+  const sourceBlockSize = Math.max(1, Math.round(blockSize / Math.min(scaleX, scaleY)));
+  const smallW = Math.max(1, Math.ceil(sourceImg.width / sourceBlockSize));
+  const smallH = Math.max(1, Math.ceil(sourceImg.height / sourceBlockSize));
+
+  const smallCanvas = document.createElement("canvas");
+  smallCanvas.width = smallW;
+  smallCanvas.height = smallH;
+  const smallCtx = smallCanvas.getContext("2d");
+  smallCtx.imageSmoothingEnabled = true;
+  smallCtx.drawImage(sourceImg, 0, 0, sourceImg.width, sourceImg.height, 0, 0, smallW, smallH);
+
+  const pixelCanvas = document.createElement("canvas");
+  pixelCanvas.width = sourceImg.width;
+  pixelCanvas.height = sourceImg.height;
+  const pixelCtx = pixelCanvas.getContext("2d");
+  pixelCtx.imageSmoothingEnabled = false;
+  pixelCtx.drawImage(smallCanvas, 0, 0, smallW, smallH, 0, 0, sourceImg.width, sourceImg.height);
+
+  return pixelCanvas;
+}
+
+function getMosaicBrushBounds(sourceImg, canvas, centerX, centerY, size) {
+  const scaleX = canvas.width / sourceImg.width;
+  const scaleY = canvas.height / sourceImg.height;
+  const sourceX = centerX / scaleX;
+  const sourceY = centerY / scaleY;
+  const radiusX = Math.max(1, (size / 2) / scaleX);
+  const radiusY = Math.max(1, (size / 2) / scaleY);
+
+  return {
+    centerX: sourceX,
+    centerY: sourceY,
+    radius: Math.max(radiusX, radiusY),
+    left: clamp(Math.floor(sourceX - radiusX), 0, sourceImg.width),
+    top: clamp(Math.floor(sourceY - radiusY), 0, sourceImg.height),
+    right: clamp(Math.ceil(sourceX + radiusX), 0, sourceImg.width),
+    bottom: clamp(Math.ceil(sourceY + radiusY), 0, sourceImg.height),
+  };
+}
+
+function expandMosaicBounds(bounds, brushBounds) {
+  if (!bounds.left && bounds.left !== 0) {
+    bounds.left = brushBounds.left;
+    bounds.top = brushBounds.top;
+    bounds.right = brushBounds.right;
+    bounds.bottom = brushBounds.bottom;
+    return;
+  }
+  bounds.left = Math.min(bounds.left, brushBounds.left);
+  bounds.top = Math.min(bounds.top, brushBounds.top);
+  bounds.right = Math.max(bounds.right, brushBounds.right);
+  bounds.bottom = Math.max(bounds.bottom, brushBounds.bottom);
+}
+
+function paintMosaicStamp(sourceImg, canvas, mosaicSource, strokeCtx, bounds, point) {
+  const brushBounds = getMosaicBrushBounds(sourceImg, canvas, point.x, point.y, point.size);
+  if (brushBounds.right <= brushBounds.left || brushBounds.bottom <= brushBounds.top) return false;
+
+  strokeCtx.save();
+  strokeCtx.beginPath();
+  strokeCtx.arc(brushBounds.centerX, brushBounds.centerY, brushBounds.radius, 0, Math.PI * 2);
+  strokeCtx.clip();
+  strokeCtx.drawImage(
+    mosaicSource,
+    brushBounds.left,
+    brushBounds.top,
+    brushBounds.right - brushBounds.left,
+    brushBounds.bottom - brushBounds.top,
+    brushBounds.left,
+    brushBounds.top,
+    brushBounds.right - brushBounds.left,
+    brushBounds.bottom - brushBounds.top
+  );
+  strokeCtx.restore();
+
+  expandMosaicBounds(bounds, brushBounds);
+  return true;
+}
+
+function paintMosaicLine(sourceImg, canvas, mosaicSource, strokeCtx, bounds, from, to, size) {
+  const dx = to.x - from.x;
+  const dy = to.y - from.y;
+  const distance = Math.hypot(dx, dy);
+  const step = Math.max(2, size * 0.25);
+  const steps = Math.max(1, Math.ceil(distance / step));
+  let painted = false;
+
+  for (let i = 1; i <= steps; i++) {
+    const t = i / steps;
+    painted = paintMosaicStamp(sourceImg, canvas, mosaicSource, strokeCtx, bounds, {
+      x: from.x + dx * t,
+      y: from.y + dy * t,
+      size,
+    }) || painted;
+  }
+
+  return painted;
+}
+
+function createMosaicStrokeObject(strokeCanvas, bounds, canvas, sourceImg) {
+  const width = bounds.right - bounds.left;
+  const height = bounds.bottom - bounds.top;
+  if (width < 1 || height < 1) return Promise.resolve(null);
+
+  const cropCanvas = document.createElement("canvas");
+  cropCanvas.width = width;
+  cropCanvas.height = height;
+  cropCanvas.getContext("2d").drawImage(
+    strokeCanvas,
+    bounds.left,
+    bounds.top,
+    width,
+    height,
+    0,
+    0,
+    width,
+    height
+  );
+
+  const scaleX = canvas.width / sourceImg.width;
+  const scaleY = canvas.height / sourceImg.height;
+  const dataUrl = cropCanvas.toDataURL("image/png");
+  return new Promise(resolve => {
+    fabric.Image.fromURL(dataUrl, (imgObj) => {
+      configureMosaicObject(imgObj).set({
+        left: bounds.left * scaleX,
+        top: bounds.top * scaleY,
+        scaleX,
+        scaleY,
+      });
+      resolve(imgObj);
+    });
+  });
+}
+
+function setupMosaicDrawing(canvas, state, sourceImg, history) {
+  let lastPoint, mosaicSource, strokeCanvas, strokeCtx, strokePreview, strokeBounds, strokePainted;
+
+  const refreshPreview = () => {
+    if (!strokePreview) return;
+    strokePreview.dirty = true;
+    canvas.requestRenderAll();
+  };
+
+  canvas.on("mouse:down", (opt) => {
+    if (state.tool !== "mosaic") return;
+    const ptr = clampPointer(canvas, canvas.getPointer(opt.e));
+    lastPoint = ptr;
+    state.drawing = true;
+    strokePainted = false;
+    strokeBounds = {};
+    mosaicSource = createMosaicSource(sourceImg, canvas, state.mosaicSize);
+
+    strokeCanvas = document.createElement("canvas");
+    strokeCanvas.width = sourceImg.width;
+    strokeCanvas.height = sourceImg.height;
+    strokeCtx = strokeCanvas.getContext("2d");
+    strokeCtx.imageSmoothingEnabled = false;
+
+    strokePreview = new fabric.Image(strokeCanvas, {
+      left: 0,
+      top: 0,
+      scaleX: canvas.width / sourceImg.width,
+      scaleY: canvas.height / sourceImg.height,
+      selectable: false,
+      evented: false,
+      excludeFromExport: true,
+      objectCaching: false,
+    });
+
+    history.locked = true;
+    canvas.add(strokePreview);
+    history.locked = false;
+
+    strokePainted = paintMosaicStamp(sourceImg, canvas, mosaicSource, strokeCtx, strokeBounds, {
+      x: ptr.x,
+      y: ptr.y,
+      size: state.mosaicSize,
+    }) || strokePainted;
+    refreshPreview();
+  });
+
+  canvas.on("mouse:move", (opt) => {
+    if (state.tool !== "mosaic" || !state.drawing || !strokeCtx || !lastPoint) return;
+    const ptr = clampPointer(canvas, canvas.getPointer(opt.e));
+    strokePainted = paintMosaicLine(
+      sourceImg,
+      canvas,
+      mosaicSource,
+      strokeCtx,
+      strokeBounds,
+      lastPoint,
+      ptr,
+      state.mosaicSize
+    ) || strokePainted;
+    lastPoint = ptr;
+    refreshPreview();
+  });
+
+  canvas.on("mouse:up", async () => {
+    if (!state.drawing || !strokePreview) return;
+
+    history.locked = true;
+    canvas.remove(strokePreview);
+    history.locked = false;
+    state.drawing = false;
+    lastPoint = null;
+
+    if (!strokePainted) {
+      strokeCanvas = null;
+      strokeCtx = null;
+      strokePreview = null;
+      mosaicSource = null;
+      canvas.renderAll();
+      return;
+    }
+
+    const mosaicObj = await createMosaicStrokeObject(strokeCanvas, strokeBounds, canvas, sourceImg);
+    mosaicSource = null;
+    strokeCanvas = null;
+    strokeCtx = null;
+    strokePreview = null;
+
+    if (mosaicObj) {
+      canvas.add(mosaicObj);
+      canvas.discardActiveObject();
+    }
+    canvas.renderAll();
+  });
+}
+
+// --- Open Paint Modal ---
+async function openPaintModal(node) {
+  injectStyles();
+  await loadFabric();
+
+  if (!node.properties) node.properties = {};
+
+  // Detect if user switched to a different image — reset saved state
+  const widget = node.widgets?.find(w => w.name === "image");
+  const currentVal = widget?.value ? String(widget.value) : "";
+  const originalImg = node.properties.paintBrushOriginal;
+  if (originalImg && currentVal !== originalImg && currentVal !== "painted_" + originalImg) {
+    // Image changed, clear old paint state
+    delete node.properties.paintBrushCanvas;
+    delete node.properties.paintBrushOriginal;
+  }
+
+  const savedState = node.properties.paintBrushCanvas;
+  const storedOriginal = node.properties.paintBrushOriginal;
+
+  // If re-editing, use the original image as background; otherwise use current
+  let bgUrl;
+  if (savedState && storedOriginal) {
+    bgUrl = `/view?filename=${encodeURIComponent(storedOriginal)}&type=input&subfolder=`;
+  } else {
+    bgUrl = getImageUrl(node);
+  }
+  if (!bgUrl) { alert("请先加载一张图片"); return; }
+
+  // Save original image name on first paint
+  if (!storedOriginal) {
+    node.properties.paintBrushOriginal = currentVal;
+  }
+
+  // Create overlay
+  const overlay = document.createElement("div");
+  overlay.className = "pb-overlay";
+
+  const maxW = window.innerWidth * 0.8;
+  const maxH = window.innerHeight * 0.75;
+
+  // Load image to get dimensions
+  const img = await new Promise((resolve, reject) => {
+    const i = new Image();
+    i.crossOrigin = "anonymous";
+    i.onload = () => resolve(i);
+    i.onerror = () => reject(new Error("图片加载失败"));
+    i.src = bgUrl;
+  });
+
+  const scale = Math.min(maxW / img.width, maxH / img.height, 1);
+  const cw = Math.round(img.width * scale);
+  const ch = Math.round(img.height * scale);
+
+  const state = { tool: "brush", color: "#ff0000", width: 4, mosaicSize: 14, drawing: false };
+
+  const toolbar = buildToolbar(state);
+  overlay.appendChild(toolbar);
+
+  const wrap = document.createElement("div");
+  wrap.className = "pb-canvas-wrap";
+  const canvasEl = document.createElement("canvas");
+  canvasEl.width = cw;
+  canvasEl.height = ch;
+  wrap.appendChild(canvasEl);
+  overlay.appendChild(wrap);
+  document.body.appendChild(overlay);
+
+  // Init Fabric canvas
+  const canvas = new fabric.Canvas(canvasEl, { width: cw, height: ch, isDrawingMode: true });
+  canvas.freeDrawingBrush.color = state.color;
+  canvas.freeDrawingBrush.width = state.width;
+
+  // Set background image (always the original)
+  await new Promise(resolve => {
+    canvas.setBackgroundImage(bgUrl, () => { canvas.renderAll(); resolve(); }, {
+      scaleX: cw / img.width, scaleY: ch / img.height, crossOrigin: "anonymous"
+    });
+  });
+
+  // Restore previous drawing objects if re-editing
+  if (savedState) {
+    await new Promise(resolve => {
+      canvas.loadFromJSON(savedState, () => {
+        normalizeMosaicObjects(canvas);
+        // Re-apply background since loadFromJSON may clear it
+        canvas.setBackgroundImage(bgUrl, () => { canvas.renderAll(); resolve(); }, {
+          scaleX: cw / img.width, scaleY: ch / img.height, crossOrigin: "anonymous"
+        });
+      });
+    });
+  }
+
+  // History
+  const history = new HistoryManager(canvas);
+  setTimeout(() => history.save(), 300);
+  canvas.on("object:added", () => history.save());
+  canvas.on("object:modified", () => history.save());
+
+  // Shape drawing
+  setupShapeDrawing(canvas, state);
+  setupMosaicDrawing(canvas, state, img, history);
+  wireToolbar(toolbar, canvas, state, history);
+
+  // Actions buttons
+  const actions = document.createElement("div");
+  actions.className = "pb-actions";
+  actions.innerHTML = `<button class="pb-cancel">取消</button><button class="pb-confirm">确认</button>`;
+  overlay.appendChild(actions);
+
+  const close = () => { canvas.dispose(); overlay.remove(); };
+  actions.querySelector(".pb-cancel").onclick = close;
+  overlay.addEventListener("keydown", (e) => { if (e.key === "Escape") close(); });
+  overlay.tabIndex = 0;
+  overlay.focus();
+
+  // Keyboard shortcuts
+  overlay.addEventListener("keydown", (e) => {
+    if (e.ctrlKey && e.key === "z" && !e.shiftKey) { e.preventDefault(); history.undo(); }
+    if (e.ctrlKey && (e.key === "Z" || (e.key === "z" && e.shiftKey))) { e.preventDefault(); history.redo(); }
+    if (e.key === "Delete" || e.key === "Backspace") {
+      const active = canvas.getActiveObject();
+      if (active) { canvas.remove(active); canvas.discardActiveObject(); history.save(); }
+    }
+  });
+
+  // Confirm - save painted image and store canvas state for re-editing
+  actions.querySelector(".pb-confirm").onclick = async () => {
+    // Save canvas objects (without background) for future re-editing
+    node.properties.paintBrushCanvas = canvas.toJSON(PB_EXPORT_PROPS);
+    node.graph?.change?.();
+    await savePaintedImage(canvas, node, img.width, img.height);
+    close();
+  };
+}
+
+// --- Save painted image ---
+async function savePaintedImage(canvas, node, origW, origH) {
+  // Export at original resolution
+  const exportCanvas = document.createElement("canvas");
+  exportCanvas.width = origW;
+  exportCanvas.height = origH;
+  const ctx = exportCanvas.getContext("2d");
+
+  const dataUrl = canvas.toDataURL({ format: "png", multiplier: origW / canvas.width });
+  const exportImg = await new Promise((resolve) => {
+    const i = new Image();
+    i.onload = () => resolve(i);
+    i.src = dataUrl;
+  });
+  ctx.drawImage(exportImg, 0, 0, origW, origH);
+
+  // Get original filename for naming (use stored original, not current painted name)
+  const origName = (node.properties.paintBrushOriginal || "").split("/").pop() || "image.png";
+  const paintedName = "painted_" + origName;
+
+  const blob = await new Promise(r => exportCanvas.toBlob(r, "image/png"));
+  const formData = new FormData();
+  formData.append("image", blob, paintedName);
+  formData.append("type", "input");
+  formData.append("overwrite", "true");
+
+  const resp = await api.fetchApi("/upload/image", { method: "POST", body: formData });
+  const data = await resp.json();
+
+  // Add to combo options so the value persists across refresh
+  const widget = node.widgets.find(w => w.name === "image");
+  if (widget) {
+    if (Array.isArray(widget.options?.values) && !widget.options.values.includes(data.name)) {
+      widget.options.values.push(data.name);
+    }
+    widget.value = data.name;
+    if (widget.callback) {
+      widget.callback(data.name);
+    }
+  }
+  // Mark graph as changed to trigger workflow auto-save
+  node.graph?.change?.();
+  app.graph.setDirtyCanvas(true, true);
+}
+
+// --- Build Toolbar ---
+function buildToolbar(state) {
+  const toolbar = document.createElement("div");
+  toolbar.className = "pb-toolbar";
+  toolbar.innerHTML = `
+    <button data-tool="brush" class="active">画笔</button>
+    <button data-tool="rect">矩形</button>
+    <button data-tool="circle">圆形</button>
+    <button data-tool="line">直线</button>
+    <button data-tool="mosaic">马赛克</button>
+    <button data-tool="eraser">橡皮擦</button>
+    <span class="pb-sep"></span>
+    <span class="pb-label">颜色</span>
+    <input type="color" class="pb-color" value="${state.color}">
+    <span class="pb-label">线宽</span>
+    <input type="range" class="pb-width" min="1" max="40" value="${state.width}">
+    <span class="pb-label">块大小</span>
+    <input type="range" class="pb-mosaic-size" min="4" max="80" value="${state.mosaicSize}">
+    <span class="pb-sep"></span>
+    <button data-action="undo">撤回</button>
+    <button data-action="clear">清空</button>
+  `;
+  return toolbar;
+}
+
+// --- Wire Toolbar ---
+function wireToolbar(toolbar, canvas, state, history) {
+  // Tool buttons
+  toolbar.querySelectorAll("[data-tool]").forEach(btn => {
+    btn.onclick = () => {
+      toolbar.querySelectorAll("[data-tool]").forEach(b => b.classList.remove("active"));
+      btn.classList.add("active");
+      state.tool = btn.dataset.tool;
+
+      if (state.tool === "brush") {
+        canvas.isDrawingMode = true;
+        canvas.selection = false;
+        canvas.defaultCursor = "default";
+        canvas.hoverCursor = "move";
+        canvas.freeDrawingBrush.color = state.color;
+        canvas.freeDrawingBrush.width = state.width;
+      } else if (state.tool === "eraser") {
+        // Eraser: click on object to delete it
+        canvas.isDrawingMode = false;
+        canvas.selection = true;
+        canvas.defaultCursor = "crosshair";
+        canvas.hoverCursor = "pointer";
+      } else if (state.tool === "mosaic") {
+        canvas.isDrawingMode = false;
+        canvas.selection = false;
+        canvas.discardActiveObject();
+        applyMosaicCursor(canvas, state);
+      } else {
+        canvas.isDrawingMode = false;
+        canvas.selection = false;
+        canvas.defaultCursor = "default";
+        canvas.hoverCursor = "move";
+      }
+    };
+  });
+
+  // Eraser: remove object on click
+  canvas.on("mouse:down", (opt) => {
+    if (state.tool !== "eraser") return;
+    const target = canvas.findTarget(opt.e);
+    if (target) {
+      canvas.remove(target);
+      canvas.discardActiveObject();
+      canvas.renderAll();
+      history.save();
+    }
+  });
+
+  // Color picker
+  toolbar.querySelector(".pb-color").oninput = (e) => {
+    state.color = e.target.value;
+    if (canvas.isDrawingMode) {
+      canvas.freeDrawingBrush.color = state.color;
+    }
+  };
+
+  // Width slider
+  toolbar.querySelector(".pb-width").oninput = (e) => {
+    state.width = parseInt(e.target.value);
+    if (canvas.isDrawingMode) {
+      canvas.freeDrawingBrush.width = state.width;
+    }
+  };
+
+  toolbar.querySelector(".pb-mosaic-size").oninput = (e) => {
+    state.mosaicSize = parseInt(e.target.value);
+    if (state.tool === "mosaic") {
+      applyMosaicCursor(canvas, state);
+    }
+  };
+
+  // Action buttons
+  toolbar.querySelector("[data-action=undo]").onclick = () => history.undo();
+  toolbar.querySelector("[data-action=clear]").onclick = () => {
+    canvas.getObjects().forEach(obj => canvas.remove(obj));
+    canvas.renderAll();
+    history.save();
+  };
+}
+
+// --- Extension Registration ---
+app.registerExtension({
+  name: "o1key.paintBrush",
+
+  // Register command for toolbar button
+  commands: [
+    {
+      id: "o1key.PaintBrush",
+      icon: "pi pi-pencil",
+      label: "画笔",
+      tooltip: "画笔",
+      function: () => {
+        const selectedNodes = app.canvas.selected_nodes;
+        const node = selectedNodes ? Object.values(selectedNodes)[0] : null;
+        if (node) openPaintModal(node);
+      }
+    }
+  ],
+
+  // Add title attribute to our toolbar button for native tooltip
+  // Hide "节点信息" button and reorder paint brush next to mask editor
+  setup() {
+    const observer = new MutationObserver(() => {
+      const toolbox = document.querySelector('[class*="selection-toolbox"], [class*="SelectionToolbox"]');
+      if (!toolbox) return;
+
+      // Hide "节点信息" button (matches by aria-label or title)
+      toolbox.querySelectorAll("button").forEach(btn => {
+        const label = btn.title || btn.getAttribute("aria-label") || btn.textContent || "";
+        if (label.includes("节点信息") || label.includes("Node Info") || label.includes("Info")) {
+          btn.style.display = "none";
+        }
+      });
+
+      // Find our paint brush button and move it next to mask editor
+      const pencilIcon = toolbox.querySelector('[class*="pi-pencil"]');
+      if (pencilIcon) {
+        const paintBtn = pencilIcon.closest("button");
+        if (paintBtn && !paintBtn.title) paintBtn.title = "画笔";
+
+        // Find mask editor button (has mask/pen-tool icon)
+        const allBtns = Array.from(toolbox.querySelectorAll("button"));
+        const maskBtn = allBtns.find(b => {
+          const cls = b.innerHTML || "";
+          return cls.includes("mask") || cls.includes("pen-tool") || cls.includes("Mask");
+        }) || allBtns[0];
+
+        if (maskBtn && paintBtn && paintBtn.previousElementSibling !== maskBtn) {
+          maskBtn.after(paintBtn);
+        }
+      }
+    });
+    observer.observe(document.body, { childList: true, subtree: true });
+  },
+
+  // Show button in toolbar when LoadImage node is selected
+  getSelectionToolboxCommands(item) {
+    if (item?.comfyClass === "LoadImage" || item?.type === "LoadImage") {
+      return ["o1key.PaintBrush"];
+    }
+    return [];
+  },
+
+  beforeRegisterNodeDef(nodeType, nodeData) {
+    if (nodeData.name !== "LoadImage") return;
+
+    const origMenu = nodeType.prototype.getExtraMenuOptions;
+    nodeType.prototype.getExtraMenuOptions = function (canvasRef, options) {
+      origMenu?.call(this, canvasRef, options);
+      options.unshift({
+        content: "画笔 (Paint)",
+        callback: () => openPaintModal(this)
+      });
+    };
+  }
+});
@@ -0,0 +1,29 @@
+import { app } from "../../../scripts/app.js";
+
+app.registerExtension({
+    name: "o1key.renameConsole",
+    async setup() {
+        const rename = () => {
+            document.querySelectorAll(".side-bar-button, [class*='sidebar'] button, .p-togglebutton, button").forEach(btn => {
+                const label = btn.getAttribute("aria-label") || "";
+                const text = btn.textContent || "";
+                if (label === "控制台" || label === "Console" || text.trim() === "控制台" || text.trim() === "Console") {
+                    if (label === "控制台" || label === "Console") {
+                        btn.setAttribute("aria-label", "日志");
+                    }
+                    const span = btn.querySelector("span");
+                    if (span && (span.textContent.trim() === "控制台" || span.textContent.trim() === "Console")) {
+                        span.textContent = "日志";
+                    } else if (!span && (btn.textContent.trim() === "控制台" || btn.textContent.trim() === "Console")) {
+                        btn.textContent = "日志";
+                    }
+                }
+            });
+        };
+
+        const observer = new MutationObserver(rename);
+        observer.observe(document.body, { childList: true, subtree: true });
+        setTimeout(rename, 1000);
+        setTimeout(rename, 3000);
+    },
+});
@@ -0,0 +1,18 @@
+import { app } from "../../../scripts/app.js";
+
+app.registerExtension({
+    name: "o1key.renameTab",
+    async setup() {
+        const rename = () => {
+            if (document.title.includes("ComfyUI")) {
+                document.title = document.title.replace("ComfyUI", "o1key");
+            }
+        };
+
+        rename();
+        new MutationObserver(rename).observe(
+            document.querySelector("title") || document.head,
+            { childList: true, subtree: true, characterData: true }
+        );
+    },
+});
@@ -0,0 +1,175 @@
+import { app } from "../../../scripts/app.js";
+
+app.registerExtension({
+    name: "o1key.restartButton",
+    async setup() {
+        function inject() {
+            if (document.querySelector("#o1k-restart-btn") && document.querySelector("#o1k-update-btn")) return;
+
+            const allBtns = document.querySelectorAll("button, .p-togglebutton, .side-bar-button");
+            let logBtn = null;
+            for (const btn of allBtns) {
+                const label = (btn.getAttribute("aria-label") || "") + (btn.textContent || "");
+                if (label.includes("日志") || label.includes("Console") || label.includes("控制台") || label.includes("Logs")) {
+                    logBtn = btn;
+                    break;
+                }
+            }
+            if (!logBtn || !logBtn.parentNode) return;
+
+            function makeButton(id, label, title, icon) {
+                const btn = logBtn.cloneNode(false);
+                btn.id = id;
+                btn.setAttribute("aria-label", label);
+                btn.title = title;
+                const logStyle = window.getComputedStyle(logBtn);
+                btn.style.display = "flex";
+                btn.style.flexDirection = "column";
+                btn.style.alignItems = "center";
+                btn.style.justifyContent = "center";
+                btn.style.gap = logStyle.gap || "4px";
+                const iconSpan = document.createElement("span");
+                iconSpan.innerHTML = icon;
+                const textSpan = document.createElement("span");
+                textSpan.textContent = label;
+                btn.append(iconSpan, textSpan);
+                return btn;
+            }
+
+            let restartBtn = document.querySelector("#o1k-restart-btn");
+            if (!restartBtn) {
+                restartBtn = makeButton("o1k-restart-btn", "重启", "重启 ComfyUI",
+                    `<svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M21 2v6h-6"/><path d="M3 12a9 9 0 0 1 15-6.7L21 8"/><path d="M3 22v-6h6"/><path d="M21 12a9 9 0 0 1-15 6.7L3 16"/></svg>`);
+                restartBtn.addEventListener("click", async () => {
+                    if (!confirm("确定要重启 ComfyUI 吗？")) return;
+                    restartBtn.style.opacity = "0.5";
+                    restartBtn.style.pointerEvents = "none";
+                    await disableExperimentalAssetApi();
+                    try { await fetch("/o1key/restart", { method: "POST" }); } catch {}
+                    pollUntilReady();
+                });
+                logBtn.parentNode.insertBefore(restartBtn, logBtn);
+            }
+
+            if (!document.querySelector("#o1k-update-btn")) {
+                const updateBtn = makeButton("o1k-update-btn", "更新", "更新 comfyui_o1key 节点包",
+                    `<svg width="18" height="18" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 3v12"/><path d="m7 10 5 5 5-5"/><path d="M4 18v3h16v-3"/></svg>`);
+                let updating = false;
+                updateBtn.addEventListener("click", async () => {
+                    if (updating) return;
+                    if (!confirm("从 origin/main 拉取 comfyui_o1key 最新版本？")) return;
+                    updating = true;
+                    updateBtn.disabled = true;
+                    updateBtn.style.opacity = "0.5";
+                    updateBtn.title = "正在更新...";
+                    try {
+                        const response = await fetch("/o1key/update", {
+                            method: "POST",
+                            headers: { "X-O1Key-Update": "1" },
+                        });
+                        const result = await response.json();
+                        if (!response.ok) throw new Error(result.error || "更新失败");
+                        if (!result.updated) {
+                            alert(`已是最新版本（${result.version}）。`);
+                        } else {
+                            const dependencies = result.requirements_changed
+                                ? "\n依赖列表已变化，请先在 ComfyUI 的 Python 环境中执行 pip install -r requirements.txt。"
+                                : "";
+                            alert(`更新完成（${result.version}）。${dependencies}\n请点击“重启”使新版本生效。`);
+                        }
+                    } catch (error) {
+                        alert(`更新失败：${error.message}`);
+                    } finally {
+                        updating = false;
+                        updateBtn.disabled = false;
+                        updateBtn.style.opacity = "";
+                        updateBtn.title = "更新 comfyui_o1key 节点包";
+                    }
+                });
+                restartBtn.after(updateBtn);
+            }
+        }
+
+        async function disableExperimentalAssetApi() {
+            if (!(await shouldDisableExperimentalAssetApi())) return;
+            try {
+                await fetch("/api/settings/Comfy.Assets.UseAssetAPI", {
+                    method: "POST",
+                    headers: { "Content-Type": "application/json" },
+                    body: JSON.stringify(false),
+                    signal: AbortSignal.timeout(2000),
+                });
+            } catch {}
+        }
+
+        async function shouldDisableExperimentalAssetApi() {
+            try {
+                const r = await fetch("/api/settings/Comfy.Assets.UseAssetAPI", {
+                    cache: "no-store",
+                    signal: AbortSignal.timeout(2000),
+                });
+                if (!r.ok || !(await r.json())) return false;
+            } catch {
+                return false;
+            }
+            return !(await fetchOk("/api/assets/seed/status", 2000));
+        }
+
+        async function fetchOk(url, timeout = 2500) {
+            try {
+                const r = await fetch(url, {
+                    cache: "no-store",
+                    signal: AbortSignal.timeout(timeout),
+                });
+                return r.ok;
+            } catch {
+                return false;
+            }
+        }
+
+        async function comfyReady() {
+            const [statsOk, modelFoldersOk] = await Promise.all([
+                fetchOk("/api/system_stats"),
+                fetchOk("/api/experiment/models"),
+            ]);
+            return statsOk && modelFoldersOk;
+        }
+
+        function pollUntilReady() {
+            let attempts = 0;
+            const maxAttempts = 80;
+            const minRestartWaitMs = 5000;
+            const startedAt = Date.now();
+            let sawUnavailable = false;
+            const interval = setInterval(async () => {
+                attempts++;
+                if (attempts > maxAttempts) { clearInterval(interval); forceReload(); return; }
+                const ready = await comfyReady();
+                if (!ready) {
+                    sawUnavailable = true;
+                    return;
+                }
+                if (!sawUnavailable && Date.now() - startedAt < minRestartWaitMs) return;
+
+                clearInterval(interval);
+                await disableExperimentalAssetApi();
+                setTimeout(forceReload, 800);
+            }, 1500);
+        }
+
+        function forceReload() {
+            window.onbeforeunload = null;
+            Object.defineProperty(BeforeUnloadEvent.prototype, "returnValue", {
+                get() { return ""; },
+                set() {}
+            });
+            location.reload();
+        }
+
+        const observer = new MutationObserver(inject);
+        observer.observe(document.body, { childList: true, subtree: true });
+        setTimeout(inject, 2000);
+        setTimeout(inject, 4000);
+        setTimeout(inject, 8000);
+    },
+});
@@ -0,0 +1,210 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+// ── marked.js 懒加载 ──────────────────────────────────────────────────────────
+let markedReady = null;
+function loadMarked() {
+    if (markedReady) return markedReady;
+    markedReady = new Promise((resolve) => {
+        if (window.marked) { resolve(window.marked); return; }
+        const s = document.createElement("script");
+        s.src = "https://cdn.jsdelivr.net/npm/marked/marked.min.js";
+        s.onload = () => resolve(window.marked);
+        s.onerror = () => resolve(null);
+        document.head.appendChild(s);
+    });
+    return markedReady;
+}
+
+// ── 节点 UI 构建 ──────────────────────────────────────────────────────────────
+function buildUI(node) {
+    if (node._spContainer) return;
+
+    const container = document.createElement("div");
+    container.style.cssText =
+        "width:100%;height:100%;box-sizing:border-box;padding:6px;" +
+        "display:flex;flex-direction:column;gap:4px;";
+
+    const toolbar = document.createElement("div");
+    toolbar.style.cssText =
+        "display:flex;justify-content:flex-end;gap:6px;align-items:center;";
+
+    const mdToggle = document.createElement("button");
+    mdToggle.textContent = "MD";
+    mdToggle.title = "切换 Markdown / 纯文本";
+    mdToggle.style.cssText =
+        "font-size:10px;padding:2px 6px;border-radius:3px;cursor:pointer;" +
+        "background:#2a5a2a;color:#ccc;border:1px solid #666;";
+
+    const copyBtn = document.createElement("button");
+    copyBtn.textContent = "复制";
+    copyBtn.style.cssText =
+        "font-size:10px;padding:2px 6px;border-radius:3px;cursor:pointer;" +
+        "background:#444;color:#ccc;border:1px solid #666;";
+
+    toolbar.appendChild(mdToggle);
+    toolbar.appendChild(copyBtn);
+
+    const isDedicatedPreview = node.comfyClass === "StreamPreview" || node.type === "StreamPreview";
+
+    const content = document.createElement("div");
+    if (isDedicatedPreview) {
+        content.style.cssText =
+            "flex:1;min-height:0;overflow:hidden;" +
+            "background:#1a1a1a;border:1px solid #444;border-radius:4px;" +
+            "padding:8px;box-sizing:border-box;font-size:13px;line-height:1.6;" +
+            "color:#ddd;white-space:pre-wrap;word-break:break-word;";
+    } else {
+        content.style.cssText =
+            "width:100%;min-height:60px;max-height:480px;overflow-y:auto;" +
+            "background:#1a1a1a;border:1px solid #444;border-radius:4px;" +
+            "padding:8px;box-sizing:border-box;font-size:13px;line-height:1.6;" +
+            "color:#ddd;white-space:pre-wrap;word-break:break-word;";
+    }
+
+    const status = document.createElement("div");
+    status.style.cssText =
+        "font-size:10px;color:#888;text-align:right;min-height:14px;";
+
+    container.appendChild(toolbar);
+    container.appendChild(content);
+    container.appendChild(status);
+
+    node._spContainer = container;
+    node._spContent = content;
+    node._spStatus = status;
+    node._spMdToggle = mdToggle;
+    node._spRawText = "";
+    node._spMarkdown = true;
+
+    mdToggle.addEventListener("click", () => {
+        node._spMarkdown = !node._spMarkdown;
+        mdToggle.style.background = node._spMarkdown ? "#2a5a2a" : "#444";
+        renderContent(node);
+    });
+
+    copyBtn.addEventListener("click", () => {
+        navigator.clipboard.writeText(node._spRawText).then(() => {
+            copyBtn.textContent = "已复制";
+            setTimeout(() => { copyBtn.textContent = "复制"; }, 1500);
+        });
+    });
+
+    const widget = node.addDOMWidget("stream_preview_widget", "preview", container, {
+        getValue() { return node._spRawText; },
+        setValue(v) { },
+    });
+    widget.computeSize = (width) => {
+        const isDedicatedPreview = node.comfyClass === "StreamPreview" || node.type === "StreamPreview";
+        if (isDedicatedPreview) {
+            const nodeHeight = node.size?.[1] ?? 320;
+            const overhead = 60;
+            return [width, Math.max(120, nodeHeight - overhead)];
+        }
+        return [width, 320];
+    };
+
+    loadMarked();
+}
+
+async function renderContent(node) {
+    const text = node._spRawText;
+    const el = node._spContent;
+    if (!text) { el.innerHTML = ""; return; }
+
+    if (node._spMarkdown) {
+        const marked = await loadMarked();
+        if (marked) {
+            el.style.whiteSpace = "normal";
+            el.innerHTML = marked.parse(text);
+        } else {
+            el.style.whiteSpace = "pre-wrap";
+            el.textContent = text;
+        }
+    } else {
+        el.style.whiteSpace = "pre-wrap";
+        el.textContent = text;
+    }
+    const isPreview = node.comfyClass === "StreamPreview" || node.type === "StreamPreview";
+    if (!isPreview) el.scrollTop = el.scrollHeight;
+}
+
+// ── 流式事件监听 ──────────────────────────────────────────────────────────────
+api.addEventListener("o1key.stream_token", (event) => {
+    const { node_id, token, done } = event.detail;
+    const node = app.graph.getNodeById(parseInt(node_id));
+    if (!node) return;
+
+    buildUI(node);
+
+    if (done) {
+        node._spStreaming = false;
+        node._spStatus.textContent = "生成完成";
+        node._spStatus.style.color = "#4a4";
+        return;
+    }
+
+    // 第一个 token 到来时清空上一次内容
+    if (!node._spStreaming) {
+        node._spStreaming = true;
+        node._spRawText = "";
+    }
+
+    node._spRawText += token;
+    node._spStatus.textContent = "生成中…";
+    node._spStatus.style.color = "#a84";
+    renderContent(node);
+});
+
+// ── 节点注册 ──────────────────────────────────────────────────────────────────
+app.registerExtension({
+    name: "comfyui_o1key.streamPreview",
+
+    async beforeRegisterNodeDef(nodeType, nodeData) {
+        if (nodeData.name !== "StreamPreview") return;
+
+        const origOnNodeCreated = nodeType.prototype.onNodeCreated;
+        nodeType.prototype.onNodeCreated = function () {
+            if (origOnNodeCreated) origOnNodeCreated.apply(this, arguments);
+            buildUI(this);
+        };
+
+        nodeType.prototype.onResize = function () {
+            this.setDirtyCanvas(true, false);
+        };
+
+        const origOnExecuted = nodeType.prototype.onExecuted;
+        nodeType.prototype.onExecuted = function (message) {
+            if (origOnExecuted) origOnExecuted.apply(this, arguments);
+            buildUI(this);
+
+            const texts = message?.text;
+            if (!texts || texts.length === 0) return;
+
+            this._spRawText = texts[0];
+            this._spStatus.textContent = "完成";
+            this._spStatus.style.color = "#4a4";
+            renderContent(this);
+            this.setDirtyCanvas(true, true);
+        };
+
+        const origOnSerialize = nodeType.prototype.onSerialize;
+        nodeType.prototype.onSerialize = function (o) {
+            if (origOnSerialize) origOnSerialize.apply(this, arguments);
+            o.sp_text = this._spRawText || "";
+            o.sp_markdown = this._spMarkdown !== false;
+        };
+
+        const origOnConfigure = nodeType.prototype.onConfigure;
+        nodeType.prototype.onConfigure = function (o) {
+            if (origOnConfigure) origOnConfigure.apply(this, arguments);
+            buildUI(this);
+            if (o.sp_text) {
+                this._spRawText = o.sp_text;
+                this._spMarkdown = o.sp_markdown !== false;
+                this._spMdToggle.style.background = this._spMarkdown ? "#2a5a2a" : "#444";
+                renderContent(this);
+            }
+        };
+    },
+});
@@ -0,0 +1,90 @@
+import { api } from "../../../scripts/api.js";
+
+function startCountdown(toast, closeBtn, seconds, accentColor) {
+    let remaining = seconds;
+    closeBtn.textContent = `× ${remaining}s`;
+
+    const interval = setInterval(() => {
+        remaining--;
+        if (remaining <= 0) {
+            clearInterval(interval);
+            toast.style.transition = "opacity 0.4s ease";
+            toast.style.opacity = "0";
+            setTimeout(() => toast.remove(), 400);
+        } else {
+            closeBtn.textContent = `× ${remaining}s`;
+        }
+    }, 1000);
+
+    closeBtn.onclick = () => {
+        clearInterval(interval);
+        toast.remove();
+    };
+}
+
+api.addEventListener("o1key.new_version", (event) => {
+    const changelog = event.detail?.changelog || [];
+
+    const style = document.createElement("style");
+    style.textContent = `
+        @keyframes o1key-slide-in {
+            from { opacity: 0; transform: translateY(16px) scale(0.97); }
+            to   { opacity: 1; transform: translateY(0) scale(1); }
+        }
+    `;
+    document.head.appendChild(style);
+
+    const toast = document.createElement("div");
+    toast.style.cssText = `
+        position: fixed;
+        bottom: 28px;
+        left: 28px;
+        background: linear-gradient(135deg, #0a1628 0%, #0d2b4e 60%, #1a4a7a 100%);
+        color: #d6eaf8;
+        border: 1px solid #2e86c1;
+        border-radius: 12px;
+        padding: 12px 16px;
+        font-size: 14px;
+        z-index: 100000;
+        box-shadow: 0 6px 24px rgba(46,134,193,0.35), 0 2px 8px rgba(0,0,0,0.5);
+        max-width: 340px;
+        animation: o1key-slide-in 0.4s cubic-bezier(.22,.68,0,1.2);
+    `;
+
+    const header = document.createElement("div");
+    header.style.cssText = `display: flex; align-items: center; justify-content: space-between;`;
+
+    const title = document.createElement("span");
+    title.textContent = "🔔 检测到有新版本发布！";
+    title.style.cssText = `font-weight: bold; font-size: 13px; color: #7fb3d3; letter-spacing: 0.5px;`;
+
+    const closeBtn = document.createElement("button");
+    closeBtn.style.cssText = `background: none; border: none; color: #7fb3d3; font-size: 13px; cursor: pointer; padding: 0; line-height: 1;`;
+
+    header.appendChild(title);
+    header.appendChild(closeBtn);
+
+    const divider = document.createElement("div");
+    divider.style.cssText = `height: 1px; background: rgba(46,134,193,0.3); margin: 8px 0;`;
+
+    toast.appendChild(header);
+    toast.appendChild(divider);
+
+    const body = document.createElement("div");
+    const items = changelog.length > 0 ? changelog : ["暂无更新说明"];
+    items.forEach(item => {
+        const line = document.createElement("div");
+        line.textContent = `• ${item}`;
+        line.style.cssText = `margin-bottom: 4px; font-size: 12px; line-height: 1.6; color: #d6eaf8;`;
+        body.appendChild(line);
+    });
+    const more = document.createElement("div");
+    more.textContent = "...";
+    more.style.cssText = `color: #7fb3d3; font-size: 12px; margin-top: 2px;`;
+    body.appendChild(more);
+
+    toast.appendChild(body);
+    document.body.appendChild(toast);
+
+    startCountdown(toast, closeBtn, 5, "#7fb3d3");
+});
@@ -0,0 +1,146 @@
+import { app } from "../../../scripts/app.js";
+import { api } from "../../../scripts/api.js";
+
+app.registerExtension({
+    name: "comfyui_o1key.videoPreview",
+
+    async beforeRegisterNodeDef(nodeType, nodeData, _app) {
+        if (nodeData.name !== "VideoPreview") return;
+
+        const origOnExecuted = nodeType.prototype.onExecuted;
+
+        nodeType.prototype.onExecuted = function (message) {
+            if (origOnExecuted) {
+                origOnExecuted.apply(this, arguments);
+            }
+
+            const videos = message?.videos;
+            if (!videos || videos.length === 0) return;
+
+            const videoInfo = videos[0];
+            const params = new URLSearchParams();
+            params.set("filename", videoInfo.filename);
+            if (videoInfo.subfolder) params.set("subfolder", videoInfo.subfolder);
+            params.set("type", videoInfo.type || "output");
+
+            const videoUrl = api.apiURL(`/view?${params.toString()}`);
+
+            // ── 首次创建 DOM 结构 ──────────────────────────────
+            if (!this._videoContainer) {
+                this._videoContainer = document.createElement("div");
+                this._videoContainer.style.cssText =
+                    "width:100%;display:flex;flex-direction:column;align-items:center;" +
+                    "padding:4px;box-sizing:border-box;";
+
+                this._videoEl = document.createElement("video");
+                this._videoEl.controls = true;
+                this._videoEl.loop = true;
+                this._videoEl.autoplay = true;
+                this._videoEl.muted = true;
+                this._videoEl.playsInline = true;
+                // 宽度铺满容器，高度由 object-fit 自适应，不限制 max-height
+                this._videoEl.style.cssText =
+                    "width:100%;display:block;border-radius:4px;" +
+                    "background:#000;object-fit:contain;";
+
+                this._videoLabel = document.createElement("div");
+                this._videoLabel.style.cssText =
+                    "font-size:10px;color:#aaa;margin-top:2px;" +
+                    "text-align:center;word-break:break-all;";
+
+                this._videoResLabel = document.createElement("div");
+                this._videoResLabel.style.cssText =
+                    "font-size:10px;color:#888;margin-top:1px;" +
+                    "text-align:center;";
+
+                this._videoContainer.appendChild(this._videoEl);
+                this._videoContainer.appendChild(this._videoLabel);
+                this._videoContainer.appendChild(this._videoResLabel);
+
+                // ── 视频元数据加载后，根据真实宽高比重新调整节点大小 ──
+                this._videoEl.addEventListener("loadedmetadata", () => {
+                    const vw = this._videoEl.videoWidth;
+                    const vh = this._videoEl.videoHeight;
+                    if (!vw || !vh) return;
+
+                    // 存储宽高比（高/宽），供 computeSize 使用
+                    this._videoAspectRatio = vh / vw;
+
+                    // 显示分辨率
+                    if (this._videoResLabel) {
+                        this._videoResLabel.textContent = `${vw} × ${vh}`;
+                    }
+
+                    // 用真实比例重新计算节点高度
+                    this._resizeToVideo();
+                });
+            }
+
+            this._videoEl.src = videoUrl;
+            this._videoLabel.textContent = videoInfo.filename;
+
+            // ── 注册 DOM Widget（仅第一次）──────────────────────
+            if (!this.widgets?.find((w) => w.name === "video_preview_widget")) {
+                const self = this;
+                const widget = this.addDOMWidget(
+                    "video_preview_widget",
+                    "div",
+                    this._videoContainer,
+                    { serialize: false, hideOnZoom: false }
+                );
+
+                // computeSize 在 LiteGraph 布局时被调用，返回 [宽, 高]
+                widget.computeSize = function (width) {
+                    const w = width ?? self.size?.[0] ?? 300;
+                    if (self._videoAspectRatio) {
+                        const innerW = Math.max(w - 16, 10); // 减去左右 padding
+                        const videoH = Math.round(innerW * self._videoAspectRatio);
+                        return [w, videoH + 40]; // +40 = 文件名 + 分辨率标签高度
+                    }
+                    // 元数据未就绪时给一个合理默认值
+                    return [w, 260];
+                };
+            }
+
+            // 初次渲染（元数据尚未加载）给出合理初始尺寸
+            if (!this._videoAspectRatio) {
+                const w = Math.max(this.size[0], 320);
+                const h = Math.max(this.size[1], 300);
+                this.setSize([w, h]);
+            }
+
+            this.setDirtyCanvas(true, true);
+        };
+
+        // ── 辅助方法：按视频真实比例自适应节点大小 ──────────────
+        nodeType.prototype._resizeToVideo = function () {
+            if (!this._videoAspectRatio) return;
+
+            const nodeWidth = Math.max(this.size[0], 320);
+            const innerW = nodeWidth - 16;
+            const videoH = Math.round(innerW * this._videoAspectRatio);
+            const labelH = 40; // 文件名 + 分辨率两行
+
+            // 节点头部 + 其他 widget 的高度
+            // LiteGraph 节点头部约 30px，每个普通 widget 约 24px
+            const NON_VIDEO_WIDGETS = (this.widgets?.filter(
+                (w) => w.name !== "video_preview_widget"
+            ).length ?? 0);
+            const headerH = 58 + NON_VIDEO_WIDGETS * 24;
+
+            const totalH = headerH + videoH + labelH;
+
+            this.setSize([nodeWidth, totalH]);
+            this.setDirtyCanvas(true, true);
+        };
+
+        // ── 节点手动缩放时同步更新视频高度 ─────────────────────
+        const origOnResize = nodeType.prototype.onResize;
+        nodeType.prototype.onResize = function (size) {
+            if (origOnResize) origOnResize.apply(this, arguments);
+            if (this._videoAspectRatio) {
+                this._resizeToVideo();
+            }
+        };
+    },
+});
@@ -1,134 +0,0 @@
-# Comfyui_o1key 更新说明
-
-## 🎉 插件已支持一键自动更新！
-
-从 v1.10.0 版本开始，插件支持自动更新功能。你只需运行更新脚本，即可轻松获取最新版本。
-
---
-
-## 📦 如何更新
-
-### Windows 用户
-
-1. 打开文件资源管理器
-2. 进入插件目录：`ComfyUI\custom_nodes\Comfyui_o1key`
-3. 双击运行 `update.bat` 文件
-4. 等待更新完成（通常只需几秒钟）
-5. 重启 ComfyUI
-
-### Linux/Mac 用户
-
-打开终端，执行以下命令：
-
-```bash
-cd ComfyUI/custom_nodes/Comfyui_o1key
-chmod +x update.sh  # 首次运行需要添加执行权限
-./update.sh
-```
-
---
-
-## ✨ 更新脚本功能
-
-✅ **自动检查更新** - 自动检测是否有新版本  
-✅ **备份配置** - 自动备份和恢复 `.config` 配置文件  
-✅ **拉取代码** - 自动从 GitHub 拉取最新代码  
-✅ **更新依赖** - 自动更新 Python 依赖包  
-✅ **显示日志** - 显示最近的更新内容  
-✅ **完善提示** - 友好的中文提示和错误处理  
-
---
-
-## 🔔 更新检查
-
-插件会在每次启动 ComfyUI 时自动检查是否有新版本：
-
- 如果发现新版本，终端会显示更新提示
- 不会影响插件加载速度
- 不会弹窗打断工作流程
- 检查失败不影响插件正常使用
-
-**终端提示示例：**
-
-```
-============================================================
-🎉 Comfyui_o1key 有新版本可用 (当前版本: v1.9.1)
-============================================================
-更新方法：
-  Windows: 双击运行 update.bat
-  Linux/Mac: 运行 ./update.sh
-或手动执行: git pull origin main
-============================================================
-```
-
---
-
-## 🛡️ 安全保障
-
- **配置安全**：更新前自动备份 `.config` 文件，更新后自动恢复
- **环境变量**：存储在系统级别的 API 密钥不受影响
- **错误处理**：更新失败不会破坏现有安装
- **回退方案**：如有问题可用 `git reset` 回退
-
---
-
-## 🔧 手动更新（备用方案）
-
-如果自动更新脚本无法使用，可以手动执行：
-
-```bash
-cd ComfyUI/custom_nodes/Comfyui_o1key
-git pull origin main
-pip install -r requirements.txt --upgrade
-```
-
---
-
-## ❓ 常见问题
-
-### Q1: 更新会覆盖我的配置吗？
-
-**不会。** 更新脚本会自动备份和恢复你的 `.config` 文件。环境变量中的 API 密钥也不受影响。
-
-### Q2: 更新失败怎么办？
-
-1. 检查网络连接是否正常
-2. 确认 Git 已正确安装
-3. 尝试手动更新（见上方"手动更新"部分）
-4. 如有未提交的修改，先备份后执行 `git reset --hard origin/main`
-
-### Q3: 更新后插件无法启动怎么办？
-
-1. 检查终端错误信息
-2. 重新运行依赖安装：`pip install -r requirements.txt --upgrade`
-3. 确认 Python 版本 ≥ 3.7
-4. 查看 [GitHub Issues](https://github.com/你的用户名/Comfyui_o1key/issues) 寻求帮助
-
-### Q4: 可以禁用启动时的更新检查吗？
-
-暂不支持配置禁用，但更新检查：
- 速度极快（<1 秒）
- 完全静默（无更新时不显示任何信息）
- 失败不影响插件加载
-
-### Q5: 如何查看当前版本？
-
-查看插件目录下的 `version.txt` 文件，或在更新时会显示当前版本号。
-
-### Q6: 多久检查一次更新？
-
-仅在 ComfyUI 启动时检查一次，不会在运行过程中反复检查。
-
---
-
-## 📮 反馈与支持
-
-如果在更新过程中遇到问题，请：
-
-1. 查看终端输出的错误信息
-2. 查阅 [GitHub Issues](https://github.com/你的用户名/Comfyui_o1key/issues)
-3. 提交新 Issue 并附上错误信息
-
---
-
-**享受自动更新带来的便利！** 🎉