Auto-Loom

Turn any web app into a narrated, self-driving MP4. The page scrolls, highlights elements, and navigates between routes while an AI voice narrates — timed to the millisecond. No manual screen recording, no fumbled Loom takes. Edit one line of narration and re-render in minutes.

How it works: ElevenLabs generates one audio clip per storyboard beat. ffprobe measures each clip's exact duration. Playwright drives the browser for exactly that long per beat while recording the window. ffmpeg muxes the final MP4 with a caption track. Audio duration is ground truth — visuals are slaved to it, so sync is exact.

This folder is the whole engine. Sibling skills are thin entry points over these same scripts:

• auto-loom-proof — Software Factory proof video: performs a build's acceptance criteria on camera, one criterion per beat