sammykumar-ComfyUI-SwissArmyKnife/web/docs/LLMStudioVideoDescribe.md

LLMStudioVideoDescribe

Extract frames from a video clip and ask an LM Studio-hosted VL model (e.g., Qwen3-VL) to describe the action.

Inputs

• base_url, model_name – LM Studio host and model identifier.
• video_path – absolute/relative path to the clip to analyze.
• sample_rate, max_duration – control how many frames are sampled.
• caption_prompt, temperature, verbose – tailor the language output and logging.

Outputs

• combined_caption – merged clip description.
• frame_captions – JSON blob of per-frame captions (mirrors combined caption in v2).
• frames_processed – integer count for sanity checks.

Usage Tips

1. Point video_path at a trimmed clip (use MediaSelection/VHS nodes to prep).
2. Adjust sample rate/duration to balance detail vs speed.
3. Feed the caption into downstream prompt or logging nodes.

Additional Resources

• Full documentation