Build log · Open source

Drop the card. Wake up to a cut.

The road-trip problem: shoot onboarding footage all day, then lose the night to editing. So we're building the fix in the open — an AI content pipeline that takes a folder of raw video, audio, music, and photos, and hands back edited cuts, captioned shorts, a thumbnail, and a tweet draft.

One hour of footage in → roughly a dozen 30-second shorts, 2–3 mid-length videos, and one long daily cut out. Everything runs locally. Nothing auto-posts — a human is always the last step.

inbox/2026-07-11/ → outbox/2026-07-11/ clip001.mp4 daily-cut.mp4 # captions burned, music ducked clip002.mp4 mid-1.mp4, mid-2.mp4 interview.wav shorts/short-N.mp4 # + 9:16 verticals photos/*.jpg thumbnail.png # hook frame + title music/*.mp3 tweet.txt # draft in the voice, never auto-posted report.json # every decision + ffmpeg command it ran

Where it actually stands

Statuses below are live from the build — same honesty rule as the trading model: what works is marked working, what doesn't isn't.

Build log

Learn it — build your own

There's a full learning portal for this now — the toolchain explained (ffmpeg, Whisper, Pillow), how the pipeline thinks stage by stage, the retention research with the actual numbers, and zero-to-first-cut setup.

→ Open the learning portal: Automated Video Production

Source: the repo is on GitHub — MIT licensed, no cloud APIs, and every run writes report.json + ffmpeg_commands.log so you can verify exactly what it did to your footage.

No roadmap promises. Watch the ledger — this page updates as the pipeline does.