Audio#Video editing#AI generation
OpenFilm turns coding agents into video agents you edit alongside
OpenFilm turns Claude Code, Codex or any coding agent into a video agent editing the same timeline you do, then exports to Final Cut Pro, Premiere or DaVinci.
Project facts
GitHub Ecosystem- License
- MIT
- Language
- TypeScript
- Stars
- 8,681
- Data checked
- 2026-11-20
Snapshot figures reflect the check date and may change over time.
Most AI video tools work like a vending machine: give a prompt, wait for the finished clip. OpenFilm takes a different route — it turns coding agents such as Claude Code or Codex into video agents and puts human and agent on the same film, live: whatever the agent changes you see as it happens, and whatever you change the agent sees too. GitHub’s API reported 8,681 stars on November 20, 2026; the repository was created on October 4, 2026.
Core features
- The coding agent is the editor: use Claude Code, Codex, any other coding agent, or the one built into the desktop app. You point at things — a clip, a stretch of the timeline, a spot in the frame, a subtitle line, a card on the canvas — and say what to change; the agent knows exactly what you mean.
- Four editors, one film: OpenFilm ships four editing surfaces, from a simple view anyone can use to a full timeline. Every editor works on the same file, so humans and agents can switch back and forth.
- Full timeline editing: trim, split, reorder and change clip speed; move, scale and restyle anything inside a shot, animate text word by word; captions, voice-over and sound all live on one timeline. Export a video, or hand the timeline to Final Cut Pro, Premiere Pro or DaVinci Resolve for finishing.
- Frame-level checks on your machine: the agent verifies every frame locally, with no Remotion or HyperFrames dependency. Each change is saved in the clip’s
overridesrather than in page code, so the agent can rewrite pages without losing your edits. - Extensible by design: plugins add new editors and connectors to outside services, while skills teach the agent a way of making a particular kind of film.
The README also explains the collaboration model: every change lands in one project file, and the editors are just different views, so switching editors loses nothing. Your manual edits are saved as per-clip overrides, so the agent can rewrite pages without overwriting them. For teams, this splits “AI drafts, human refines, post finishes” into three roles working on the same timeline.
Typical use cases
- Teams that already use coding agents: keep your tool;
npx skills addinstalls the skill, and your existing agent starts making video. - Films that need human sign-off: AI drafts, you point and ask for changes — a fit for marketing clips, product demos and social shorts.
- Professional finishing pipelines: OpenFilm handles assembly, then the timeline goes to FCP, Premiere or DaVinci for grading and mixing. Paired with speech capabilities such as Google’s on-device speech, it also covers narration.
Quick start
The README’s skill install (needs Node 22.2+ and ffmpeg):
npx skills add openfilm/openfilm
Then ask your agent for a film. Alternatively use the desktop app, which ships the built-in agent and everything needed to render, or the openfilm npm package for integration. The openfilm/examples repository holds complete films as folders you can open, change and render.
Summary
OpenFilm turns “AI video” from a one-shot generation into a collaborative workflow: one project file, several editors, a coding agent as the executor and a human as the director. Its limits are clear too — output quality depends on the agent and the footage, complex effects still belong in post, and local frame-by-frame checks demand a capable machine. GitHub’s API reported TypeScript, MIT, 8,681 stars, an unarchived repository and a push on November 20, 2026.
This entry is based on the README and project site; we did not install it or cut a film. Test the editors and export compatibility inside your own post pipeline before relying on them.