Jony
ba920f2b66
Replace the prior release tree with the current plugin, frontend, tests, and documentation. Document retired node IDs and the public Gitea update source.
232 lines
8.3 KiB
Python
232 lines
8.3 KiB
Python
"""Media validation helpers for MiniMax-H3 reference inputs."""
|
||
|
||
import io
|
||
import os
|
||
from typing import Any, Dict, Iterable
|
||
|
||
|
||
MB = 1024 * 1024
|
||
MIN_MEDIA_DIMENSION = 256
|
||
MAX_MEDIA_DIMENSION = 5760
|
||
MIN_MEDIA_RATIO = 0.4
|
||
MAX_MEDIA_RATIO = 2.5
|
||
|
||
MAX_IMAGE_BYTES = 30 * MB
|
||
MAX_REFERENCE_IMAGES = 9
|
||
|
||
MAX_VIDEO_BYTES = 50 * MB
|
||
MAX_REFERENCE_VIDEOS = 3
|
||
MIN_REFERENCE_DURATION = 2.0
|
||
MAX_REFERENCE_DURATION = 15.0
|
||
MAX_TOTAL_VIDEO_DURATION = 15.0
|
||
MIN_VIDEO_FPS = 23.976
|
||
MAX_VIDEO_FPS = 60.0
|
||
|
||
MAX_AUDIO_BYTES = 15 * MB
|
||
MAX_REFERENCE_AUDIOS = 3
|
||
MAX_TOTAL_AUDIO_DURATION = 15.0
|
||
|
||
|
||
def _validate_dimensions(width: int, height: int, label: str) -> None:
|
||
if not (
|
||
MIN_MEDIA_DIMENSION <= width <= MAX_MEDIA_DIMENSION
|
||
and MIN_MEDIA_DIMENSION <= height <= MAX_MEDIA_DIMENSION
|
||
):
|
||
raise ValueError(
|
||
f"{label}宽高必须均在 {MIN_MEDIA_DIMENSION}～{MAX_MEDIA_DIMENSION}px，"
|
||
f"当前为 {width}x{height}。"
|
||
)
|
||
ratio = width / height
|
||
if not MIN_MEDIA_RATIO <= ratio <= MAX_MEDIA_RATIO:
|
||
raise ValueError(
|
||
f"{label}宽高比必须在 0.4～2.5，当前为 {ratio:.3f}（{width}:{height}）。"
|
||
)
|
||
|
||
|
||
def validate_image(image, label: str = "图片") -> Dict[str, Any]:
|
||
"""Validate the exact PNG bytes sent by the shared uploader."""
|
||
width, height = image.size
|
||
_validate_dimensions(int(width), int(height), label)
|
||
|
||
buffer = io.BytesIO()
|
||
image.save(buffer, format="PNG")
|
||
size = buffer.tell()
|
||
if size > MAX_IMAGE_BYTES:
|
||
raise ValueError(
|
||
f"{label}转为 PNG 后不能超过 30MB，当前为 {size / MB:.2f}MB。"
|
||
)
|
||
return {"width": width, "height": height, "size": size}
|
||
|
||
|
||
def _get_video_source(video):
|
||
if hasattr(video, "get_stream_source"):
|
||
source = video.get_stream_source()
|
||
elif isinstance(video, dict):
|
||
source = (
|
||
video.get("video")
|
||
or video.get("path")
|
||
or video.get("file")
|
||
or video.get("filename")
|
||
or video.get("source_path")
|
||
)
|
||
elif isinstance(video, (str, os.PathLike, io.BytesIO)):
|
||
source = video
|
||
else:
|
||
source = None
|
||
for attr in ("source_path", "path", "video", "file", "filename"):
|
||
if hasattr(video, attr):
|
||
source = getattr(video, attr)
|
||
if source:
|
||
break
|
||
if source is None:
|
||
raise ValueError("无法获取参考视频数据。")
|
||
return source
|
||
|
||
|
||
def _source_for_probe(source):
|
||
if isinstance(source, io.BytesIO):
|
||
data = source.getvalue()
|
||
return io.BytesIO(data), len(data)
|
||
try:
|
||
path = os.fspath(source)
|
||
except TypeError:
|
||
raise ValueError(f"无法读取参考视频来源：{type(source).__name__}") from None
|
||
if not os.path.isfile(path):
|
||
raise ValueError(f"参考视频文件不存在：{path}")
|
||
return path, os.path.getsize(path)
|
||
|
||
|
||
def probe_video(video) -> Dict[str, Any]:
|
||
"""Inspect a ComfyUI VIDEO using PyAV without transcoding it."""
|
||
try:
|
||
import av
|
||
except ImportError:
|
||
raise RuntimeError("当前 ComfyUI 环境缺少 PyAV，无法校验 MiniMax H3 参考视频。") from None
|
||
|
||
source = _get_video_source(video)
|
||
probe_source, size = _source_for_probe(source)
|
||
try:
|
||
container = av.open(probe_source)
|
||
except Exception as exc:
|
||
raise ValueError(f"无法读取参考视频：{exc}") from None
|
||
|
||
try:
|
||
video_streams = list(container.streams.video)
|
||
if not video_streams:
|
||
raise ValueError("参考视频不包含视频轨道。")
|
||
stream = video_streams[0]
|
||
width = int(stream.codec_context.width or 0)
|
||
height = int(stream.codec_context.height or 0)
|
||
video_codec = str(stream.codec_context.name or "").lower()
|
||
rate = stream.average_rate or stream.base_rate or stream.guessed_rate
|
||
fps = float(rate) if rate else 0.0
|
||
|
||
duration = None
|
||
if stream.duration is not None and stream.time_base is not None:
|
||
duration = float(stream.duration * stream.time_base)
|
||
elif container.duration is not None:
|
||
duration = float(container.duration / av.time_base)
|
||
duration = float(duration or 0.0)
|
||
|
||
audio_codecs = {
|
||
str(audio_stream.codec_context.name or "").lower()
|
||
for audio_stream in container.streams.audio
|
||
}
|
||
format_names = {
|
||
name.strip().lower()
|
||
for name in str(container.format.name or "").split(",")
|
||
if name.strip()
|
||
}
|
||
finally:
|
||
container.close()
|
||
|
||
return {
|
||
"size": size,
|
||
"width": width,
|
||
"height": height,
|
||
"duration": duration,
|
||
"fps": fps,
|
||
"video_codec": video_codec,
|
||
"audio_codecs": audio_codecs,
|
||
"format_names": format_names,
|
||
}
|
||
|
||
|
||
def validate_video_info(info: Dict[str, Any], label: str = "参考视频") -> None:
|
||
size = int(info.get("size") or 0)
|
||
if size <= 0:
|
||
raise ValueError(f"{label}文件为空。")
|
||
if size > MAX_VIDEO_BYTES:
|
||
raise ValueError(f"{label}不能超过 50MB，当前为 {size / MB:.2f}MB。")
|
||
|
||
format_names = set(info.get("format_names") or ())
|
||
if not format_names.intersection({"mp4", "mov"}):
|
||
raise ValueError(f"{label}容器仅支持 MP4/MOV，当前为 {sorted(format_names) or '未知'}。")
|
||
|
||
duration = float(info.get("duration") or 0.0)
|
||
if not MIN_REFERENCE_DURATION <= duration <= MAX_REFERENCE_DURATION:
|
||
raise ValueError(f"{label}时长必须为 2～15 秒，当前为 {duration:.3f} 秒。")
|
||
|
||
fps = float(info.get("fps") or 0.0)
|
||
if not MIN_VIDEO_FPS <= fps <= MAX_VIDEO_FPS:
|
||
raise ValueError(f"{label}帧率必须为 23.976～60 FPS，当前为 {fps:.3f} FPS。")
|
||
|
||
def validate_reference_videos(videos: Iterable[Any]) -> list[Dict[str, Any]]:
|
||
videos = list(videos)
|
||
if len(videos) > MAX_REFERENCE_VIDEOS:
|
||
raise ValueError(f"参考视频最多 {MAX_REFERENCE_VIDEOS} 个，当前为 {len(videos)} 个。")
|
||
infos = []
|
||
for index, video in enumerate(videos, start=1):
|
||
info = probe_video(video)
|
||
validate_video_info(info, f"参考视频{index}")
|
||
infos.append(info)
|
||
total_duration = sum(float(info["duration"]) for info in infos)
|
||
if total_duration > MAX_TOTAL_VIDEO_DURATION:
|
||
raise ValueError(f"参考视频总时长不能超过 15 秒，当前为 {total_duration:.3f} 秒。")
|
||
return infos
|
||
|
||
|
||
def inspect_audio(audio, label: str = "参考音频") -> Dict[str, Any]:
|
||
if not isinstance(audio, dict):
|
||
raise ValueError(f"{label}数据格式无效。")
|
||
waveform = audio.get("waveform")
|
||
sample_rate = int(audio.get("sample_rate") or 0)
|
||
if waveform is None or sample_rate <= 0:
|
||
raise ValueError(f"{label}缺少 waveform 或 sample_rate。")
|
||
|
||
shape = tuple(int(v) for v in waveform.shape)
|
||
if not shape:
|
||
raise ValueError(f"{label}波形为空。")
|
||
if len(shape) == 3 and shape[0] != 1:
|
||
raise ValueError(f"{label}仅支持一个音频批次，当前批次为 {shape[0]}。")
|
||
samples = shape[-1]
|
||
if samples <= 0:
|
||
raise ValueError(f"{label}波形为空。")
|
||
duration = samples / sample_rate
|
||
|
||
# The shared uploader currently downmixes to mono 16-bit PCM WAV.
|
||
encoded_size = 44 + samples * 2
|
||
if encoded_size > MAX_AUDIO_BYTES:
|
||
raise ValueError(
|
||
f"{label}编码为 WAV 后不能超过 15MB，预计为 {encoded_size / MB:.2f}MB。"
|
||
)
|
||
if not MIN_REFERENCE_DURATION <= duration <= MAX_REFERENCE_DURATION:
|
||
raise ValueError(f"{label}时长必须为 2～15 秒，当前为 {duration:.3f} 秒。")
|
||
return {
|
||
"sample_rate": sample_rate,
|
||
"samples": samples,
|
||
"duration": duration,
|
||
"encoded_size": encoded_size,
|
||
}
|
||
|
||
|
||
def validate_reference_audios(audios: Iterable[Any]) -> list[Dict[str, Any]]:
|
||
audios = list(audios)
|
||
if len(audios) > MAX_REFERENCE_AUDIOS:
|
||
raise ValueError(f"参考音频最多 {MAX_REFERENCE_AUDIOS} 个，当前为 {len(audios)} 个。")
|
||
infos = [inspect_audio(audio, f"参考音频{index}") for index, audio in enumerate(audios, 1)]
|
||
total_duration = sum(float(info["duration"]) for info in infos)
|
||
if total_duration > MAX_TOTAL_AUDIO_DURATION:
|
||
raise ValueError(f"参考音频总时长不能超过 15 秒，当前为 {total_duration:.3f} 秒。")
|
||
return infos
|