video-analysislisted

# Video Analysis (Frames + Summary) Use this skill when the user wants visual analysis of a video — keyframe extraction, per-frame description, and/or a structured summary. For the transcript step, see the `audio-analysis` skill — it handles long files via chunking automatically. ## Pick a flow before running anything Two flows produce analysis for a video. Default to (A); only branch to (B) when the user's intent fits. **(A) Agent-driven flow (default, FREE).** You — the agent — extract keyframes, run your own vision on each, and write structured `FrameDescription` / `VideoSummary` via `libi.analysis_save_*`. Strengths: free, gives per-frame bboxes (needed for tracking and character catalog), gives you control over what to look for. Weaknesses: per-frame independent (weak cross-shot continuity), can't hear audio (no structured music/SFX), slow on long videos. **(B) Model-driven script flow (PAID, opt-in).** Run `libi.extra_analysis_model`. A single fal.ai → Gemini 2.5 Pro call returns a structured `Script` (shot list with camera/lighting/mood/dialogue per shot, structured music & sound-design, overall style, pacing). Strengths: hears audio, sees the whole video at once (real shot boundaries, real music cues), one fast call. Weaknesses: paid (fal credits), no per-frame bboxes, no word-level transcript timing. > **`focus` parameter on (B).** `libi.extra_analysis_model` accepts `focus: "script" | "captions"` (default `"script"`). `focus: "captions"` runs a caption-focused