About

Add emphasis motion to a flat talking-head video: dramatic hard-cut punch-zooms on emphasised words and listicles, and subtle Ken Burns slow-pushes on insightful lines. Transcript-driven, ffmpeg-based, non-destructive.

Built for vertical 9:16 social clips — LinkedIn hot-takes, TikToks, Shorts — where a flat talking-head needs energy without a manual editing timeline.

You give it a video and a small zoom map (a JSON list of moments). A hard-cut zoom reads as intentional only when it lands exactly on the word's onset — timing is the whole game.

What it does

• Punch — hard cut in on a single emphasised word, hold, hard cut out
• Ratchet — a listicle hard-cuts up a level per item, then cuts back
• Ken Burns — a slow continuous push on an insight or thesis line (the only smooth move)
• Overlapping events blend via a max envelope, so releases lift instead of snapping down
• Audio is stream-copied and output is +faststart for social; re-render after edits in ~5× realtime

How to use it

1. Transcribe for word timings, then write a zoom map placing each t on a word's start.
2. Render: python3 scripts/zoom_lib.py render input.mp4 output.mp4 map.json.
3. Prove the difference with a synced split-screen via scripts/make_comparison.py.