🖼️ 获取片段图片和音频

SplitVideoByFrames chops a video into a list of clips - but a Python list isn't something you can wire around ComfyUI's canvas. This node is the bridge: give it the clips list and an index, and it hands you that one clip's frames as a real IMAGE tensor, its audio track, and how many frames are in it. "🖼️ 获取片段图片和音频". Think of it as the "unpack one chunk" step that makes the whole video-understanding pipeline work.

How it works

The clips input is the LIST that SplitVideoByFrames outputs - a list of dicts, each holding {"frames": tensor, "audio": dict}. This node just indexes into it: clips[index], then returns the pieces. frames comes out as a (T, H, W, C) IMAGE batch (T frames, ready for a video-aware model or per-frame processing), audio comes out as a standard ComfyUI AUDIO dict (the full clip soundtrack), and num_frames is the frame count for logic. It does bounds-checking and raises an IndexError if your index is out of range.

Inputs and outputs

• index (INT, default 0) - which clip to pull. clips[0] is the first one.
• clips (LIST) - the list from SplitVideoByFrames.
• Outputs: images (IMAGE), audio (AUDIO), num_frames (INT).

The natural workflow

The full loop in this pack is a nice demonstration of how these nodes compose:

1. SplitVideoByFrames - long video in, N clips out.
2. GetVideoClipByIndex - pick clip #k, get its frames and audio.
3. Process those frames however you like - img2img, a vision LLM, captioning.
4. Optionally re-encode the result with CloudImagesToVideoAndUpload (pass the same audio back in) and push it to Qiniu.

That's a "video in, remixed video URL out" pipeline assembled from three nodes with no Python written. You drive the "which clip" choice manually by setting index, or wire it to a counter if you're iterating.

Installing it

Standard pack install - ComfyUI Manager → search comfyUI_LLM, or:

cd ComfyUI/custom_nodes
git clone https://github.com/XieJunchen/comfyUI_LLM
# restart ComfyUI

No extra dependencies beyond the pack's base requirements.

Where people get burned

• clips is opaque on the canvas. You can't inspect a LIST the way you can inspect an IMAGE preview. You'll know the clip count from SplitVideoByFrames' num_clips output - wire that to a text display so you don't guess indices blind.
• Index out of range → hard error. It raises rather than returning empty. Check num_clips first, especially if you're iterating with a changing video.
• num_frames is per-clip, not total. Easy to misread if you're computing progress across the whole video.

For any workflow that splits a video to make it digestible, this is the node that makes the split usable. Small, no magic, and it does exactly what the name says.