Prompt To Video

prompt-to-video

[](LICENSE)

A [Claude Code](https://claude.com/claude-code) Skill that turns a plain-language prompt into a narrated, captioned, ready-to-post video — using [Remotion](https://remotion.dev) to render React components to MP4.

Tell Claude what you want ("make a TikTok video explaining gradient descent") and it writes the script, generates AI narration, aligns frame-accurate captions, builds the animated scenes, previews them, and renders the final video — end to end, in one conversation.

↑ 4.5s clip from examples/gradient-descent.mp4 — full video, script, and prompt below

Why this exists

Writing video code is normally flying blind: you write a scene, wait minutes for a full render, and only then find out the camera jitters or the text runs off-screen. And narrated explainer videos have a second, harder problem — every time you tweak a line of narration, every animation timed against the old audio drifts out of sync.

This skill closes both loops:

• See before you render. A sparse-frame preview renders a handful of stills in seconds, so Claude (and you) can look at the actual output before committing to a multi-minute render.
• Animation follows the words, not a frame number. Narration timing comes from forced alignment against the real audio, so scene animations are anchored to phrases ("the moment it says 'multiply'") instead of hardcoded frame counts. Re-record the narration and every animation re-times itself automatically.

Features

• Script-driven pipeline — one script.json file is the only thing you hand-edit; audio, captions, and cue timings are all generated from it and kept in sync automatically