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

MediaDescribe

Describe images or video clips with Google's Gemini API, combining MediaSelection inputs, optional overrides, and caching so downstream prompt builders get structured data.

Inputs

• media_source / media_type / path & upload fields – choose whether to pull a random asset, a fixed path, or an uploaded file.
• media_info, resize_width, resize_height, max_duration, fps – keep Gemini's request aligned with the workflow resolution/duration.
• gemini_options from GeminiUtilOptions plus optional overrides and prompt prefix fields to steer the narration.
• seed, prompt_style, fallback_mode controls for deterministic captions across batch/exploration runs.

Outputs

• all_media_describe_data – master JSON blob with every paragraph and metadata point.
• raw_llm_json – untouched Gemini response (for auditing/debugging).
• positive_prompt_json & positive_prompt – structured and flattened prompt text ready for Control Panel or generation.
• prompt_request – the exact instructions sent to Gemini, plus inferred height/width.

Usage Tips

1. Connect MediaSelection (or VHS loaders) to feed frames plus metadata.
2. Chain GeminiUtilOptions and optional MediaDescribeOverrides into their respective ports.
3. Send the outputs to downstream nodes such as ShowText, ControlPanel*, or filename/build nodes.

Additional Resources

• Full documentation