feat: 实现 K3 图生视频 / 多镜头后端请求

- 模型名动态拼接(模式/时长/音频),不暴露于前端
- 模式前端重命名为 标准/专家/4K;4K 模式不拼接 voice 后缀
- 起始帧改为必填,仅作图生视频;移除尾帧参数
- 图片预处理:宽高比校验、等比缩放(最小 300px / 最大 10MB)
- 多镜头模式:metadata 携带 multi_shot/shot_type/multi_prompt
- 多镜头前端选项重命名为 禁用/N个故事板
- 多镜头提示词为空时 fallback 为空格,规避 API prompt is required
This commit is contained in:
Jony
2026-04-24 23:18:27 +08:00
parent dbf6bdfcc8
commit 53384f3820
+304 -35
View File
@@ -1,9 +1,19 @@
"""
K3 图生视频 自研节点(方案A:经典 INPUT_TYPES 前端壳子
复刻 Kling 3.0 Video 的前端参数,多镜头通过下拉选择控制分镜数量
后端请求尚未实现。
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 re
import aiohttp
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
try:
from comfy_api.latest import InputImpl
import folder_paths
@@ -14,51 +24,123 @@ except Exception:
# ── 常量 ──────────────────────────────────────────────────────────────────────
_MODEL_BASE = "kling-v3" # 动态拼接为 kling-v3-{模式}-{时长}s-{voice}
_MODES = ["标准", "专家", "4K"]
_MODE_MAP = {"标准": "std", "专家": "pro", "4K": "4k"}
_MULTI_SHOT_OPTIONS = [
"disabled",
"1 storyboard",
"2 storyboards",
"3 storyboards",
"4 storyboards",
"5 storyboards",
"6 storyboards",
"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",
]
_MODELS = ["kling-v3"]
_MODES = ["std", "pro", "4k"]
_ENDPOINT_CREATE = "/v1/video/generations"
_ENDPOINT_STATUS = "/v1/video/generations/{task_id}"
_POLL_INIT = 3
_POLL_MAX = 15
# ── 工具函数 ───────────────────────────────────────────────────────────────────
def _get_video_dir() -> str:
if _FOLDER_PATHS_OK:
base = folder_paths.get_output_directory()
else:
plugin = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
base = os.path.join(os.path.dirname(os.path.dirname(plugin)), "output")
d = os.path.join(base, "video")
os.makedirs(d, exist_ok=True)
return d
def _next_counter(directory: str, prefix: str) -> int:
pattern = re.compile(rf"^{re.escape(prefix)}_(\d+)")
max_n = 0
if os.path.exists(directory):
for f in os.listdir(directory):
m = pattern.match(f)
if m:
max_n = max(max_n, int(m.group(1)))
return max_n + 1
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(前端壳子,后端待实现)"""
"""K3 图生视频 自研"""
@classmethod
def INPUT_TYPES(cls):
required = {
# 多镜头模式选择
"多镜头": (_MULTI_SHOT_OPTIONS, {
"default": "disabled",
"tooltip": "disabled:单段模式;N storyboards:启用 N 段分镜。",
"default": "禁用",
"tooltip": "禁用:单段模式;N个故事板:启用 N 段分镜。",
}),
# 单段模式字段(多镜头 disabled 时使用)
"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,
"tooltip": "单段模式时长(秒);多镜头模式下由各分镜时长决定。",
}),
# 通用参数
"时长": ([5, 10, 15], {"default": 5}),
"生成音频": (["关闭", "打开"], {"default": "关闭"}),
"": (_MODELS, {"default": "kling-v3"}),
"模式": (_MODES, {"default": "std"}),
"": (_MODES, {"default": "标准"}),
"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。",
}),
}
# 分镜字段放最后,按 分镜N_提示词 / 分镜N_时长 交替排列
for i in range(1, 7):
required[f"分镜{i}_提示词"] = ("STRING", {
"multiline": True, "default": "",
@@ -70,22 +152,209 @@ class K3Video:
"tooltip": f"{i} 段分镜时长(秒)。",
})
optional = {
"起始帧": ("IMAGE",),
"尾帧": ("IMAGE",),
}
return {"required": required, "optional": optional}
return {"required": required}
RETURN_TYPES = ("VIDEO",)
RETURN_NAMES = ("视频",)
FUNCTION = "generate"
CATEGORY = "comfyui_o1key/KVideo"
async def generate(self, 多镜头, 提示词, 负向提示词, 时长,
生成音频, 模型, 模式, seed,
起始帧=None, 尾帧=None, **kwargs):
raise NotImplementedError("K3 图生视频 自研:后端尚未实现。")
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, seed, **kwargs):
api_key = get_api_key_or_raise()
base_url = get_api_base_url()
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["generate_audio"] = True
# ── 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 ──────────────────────────────────────────────────
video_dir = _get_video_dir()
counter = _next_counter(video_dir, "k3")
save_path = os.path.join(video_dir, f"k3_{counter:05d}.mp4")
connector = aiohttp.TCPConnector(ssl=False, force_close=True)
async with aiohttp.ClientSession(connector=connector) as session:
# 1. 提交
_stage("submitting")
create_url = f"{base_url}{_ENDPOINT_CREATE}"
async with session.post(create_url, json=body, 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"K3 提交失败 ({resp.status}): {msg}")
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:
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 = (data.get("status") or sr.get("status") or "").lower()
pct_raw = data.get("progress", 0)
try:
pct = int(str(pct_raw).rstrip("%").strip())
except (ValueError, AttributeError):
pct = 0
print(f"[K3 {tag}] 生成中 {pct}%")
_progress(pct)
if status in ("success", "completed", "done", "finished", "succeed"):
video_url = (
data.get("video_url")
or data.get("result_url")
or data.get("url")
or (data.get("result", {}) or {}).get("url")
or sr.get("video_url")
or sr.get("url")
)
break
if status in ("failed", "fail"):
err_info = data.get("error") or sr.get("error") or {}
err_msg = (err_info.get("message", "未知错误")
if isinstance(err_info, dict) else str(err_info))
raise RuntimeError(f"K3 生成失败:{err_msg}")
await asyncio.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
_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.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):
f.write(chunk)
_stage("done")
if _FOLDER_PATHS_OK:
return (InputImpl.VideoFromFile(save_path),)
return (save_path,)
# ── 节点注册 ──────────────────────────────────────────────────────────────────