凤希AI - 视频加载

FxAiVideoLoad ("凤希AI - 视频加载") loads a video from a folder and hands you its frames as tensors plus a pile of useful metadata. What's interesting isn't what it does - every video pack has a loader - but how it does it: no PyAV, no OpenCV, no moviepy. It shells out to ffprobe for metadata and ffmpeg for raw-frame decoding. The only real dependency is that you have ffmpeg on your PATH.

What you get

Point it at a folder path and a video index (the videos are sorted by filename, so 0 is the first), and the outputs are:

• 视频帧序列(视频格式) - IMAGE, the decoded frames as a single video-format tensor. Feed this straight into the pack's V3/V4 generators or a video-aware sampler.
• 所有帧图片序列(图片格式) - IMAGE, the same frames in image-format. In practice both outputs are currently the same tensor (don't read anything into the naming), but the split exists so the pack can diverge them later.
• 当前视频路径 - STRING, the full path, handy for preview or merge steps.
• 总帧数 - INT.
• 帧率(FPS) - INT (rounded).
• 分辨率 - STRING like "1920x1080".

Two time inputs make it a slicer too: 起始时间 and 结束时间 (both FLOAT, seconds). Set 结束时间 to 0 (the default) and it decodes to the end of the file. Set a range and you get just that window - useful for grabbing a few seconds to use as a continuity reference instead of loading a whole 10-minute file into RAM.

How it works

ffprobe reads the video stream's frame rate, dimensions, and duration; then ffmpeg decodes raw rgb24 frames piped to stdout, which the node reads frame-by-frame and converts to ComfyUI tensors. That means decoding is CPU-side and raw - a big 1080p file will chew through RAM while it loads, so the time-slice inputs aren't just a nicety, they're a memory valve.

The gotchas

• ffmpeg must be installed and findable. No ffmpeg, no frames - the node raises a clean "FFprobe 无法读取视频" error, so at least it's loud about it. On Windows, make sure ffmpeg is on PATH rather than just sitting in a folder somewhere.
• The folder must contain supported extensions (.mp4 .avi .mov .mkv .flv .wmv .mpeg .mpg .webm) and the 视频索引 must be in range, or you get a RuntimeError. The folder path is absolute - unlike the pack's other video nodes, this one doesn't resolve through the fxai/video base, so paste the full path or wire it from FxAiVideoManager's 文件夹路径 output.

Install

Part of fxai-toolkit:

cd ComfyUI/custom_nodes
git clone https://github.com/fxai666/fxai-toolkit

Restart ComfyUI, or use ComfyUI Manager and search "fxai-toolkit". No Python video deps to fight - just system ffmpeg, which the rest of the pack's video/audio nodes need anyway.