Set it up once

Unzip the folder, open your agent in it, and paste the setup prompt from the README. The agent runs node tools/doctor.mjs --fix, which checks Node.js, pnpm, dependencies, ffmpeg, Python, the Whisper and voice environment, and the render browser. It installs what it can, prints the exact command for anything it can't, and finishes with a real rendered test frame.

Type one line

make a reel: 3 sleep mistakes that ruin your mornings
reel banao: 3 sleep mistakes that ruin your mornings

Want a style? Put it in the line: make a reel: topic, pixel avatar, dark preset.

The seven stages

01. Research

It opens real sources. A facts check rejects any claim without an opened source, and any on-screen number that is not in the research.

Command: pnpm run facts. Output: research/sources.md.

02. Two scripts

Two different angles, each with the spoken script, chapters, what you will see in each one, the visual direction, sources and voice samples.

Command: pnpm run gate. Output: script/variant-a.md.

03. You approve

The only stop. Pick A or B, pick a voice, or ask for changes. Nothing is voiced or built before your yes.

Command: pnpm run approve. Output: gate.md.

04. Voice, music, captions

A natural voice at around four words a second, a music bed from your own folder, and captions synced to every word.

Command: pnpm run sync. Output: captions.words.json.

05. Scenes

Planned on the real voice timings, one code file per scene, with the card on screen changing state on the spoken word.

Command: pnpm run plan. Output: scenes/01-hook.tsx.

06. Self-review

At most three rounds. Frames are checked with Instagram's no-go zones drawn on, plus spot checks, preview clips and motion rhythm, then fixed.

Command: pnpm run review. Output: review/round-1/.

07. Final

Your 1080 × 1920 video, plus a caption, hashtags and the source links, all in one folder you own.

Command: pnpm run final. Output: final/my-reel.mp4.

What stays on your machine

Everything about a reel lives in projects/<id>/: the brief, sources, both scripts, your approval in your own words, a decision log, the review rounds and the final video. Stop any time, and later the agent picks it up with pnpm run status -- <id>.

Voice, music and captions

Edge neural voices are free and need internet. Options include warm, casual, Indian English and Hindi voices. Speech is mastered to -16 LUFS with pauses capped at 0.18 seconds.
Kokoro runs offline with --provider kokoro.
Your own recording works in any format with --provider own.
Music comes from your own music/ folder. The bed loops, ducks under the voice and fades out.
Captions are timed locally by Whisper and aligned to your exact text.

Check the terms of your voice service and music tracks before commercial use.

Get the kit for $16 once