OpenFilm: Tell Your Coding Agent "Make Me a Video" — Then Edit It Like a Real Editor

Everyone can generate a video clip now. Almost nobody can edit what the AI made. OpenFilm — an open-source project that hit 975 stars within two weeks of launch — attacks exactly that gap: your coding agent writes the film as web pages, and you get a real timeline editor to fix every frame of it.

The dirty secret of AI video

Generate a clip with today's tools and you get one shot. Want to change a word on screen, trim two seconds, or restyle a lower-third? You regenerate and pray. The output is a lottery ticket: every tweak means rolling the dice again, with no guarantee the parts you liked survive.

OpenFilm's answer is disarmingly simple: make the film out of something your agent already understands — the web. Your coding agent (Claude Code, Codex, anything that writes code) already knows HTML, CSS, and JavaScript cold. So OpenFilm has it write the entire film as web pages, then hands you a genuine editor to refine the result.

How it actually works: one file, one function, one editor

One HTML file. The whole film lives in film.html — animated web pages, footage, pictures, and sound cut together on tracks, styled with CSS. Nothing to learn; it's the stack your agent uses every day.
One function per shot. Each shot is a web page that can draw any moment through a frame(t) function — three.js, WebGL, canvas, SVG, CSS animations, GSAP, video elements. No heavy framework like Remotion standing between you and the pixels.
One real editor. Studio is a local app running in your browser: move, scale, and rotate any element; change text, color, and style; place, crop, and restyle clips; trim and cut on a timeline. The things that are impossible with generated clips become drag-and-drop.

The architecture matters because nothing is a black box. When something looks wrong, you can open the file and see exactly why — then fix it like any other code.

Install and first run

You need Node 22.2+ and ffmpeg (macOS, Windows, Linux all work); the first run downloads a headless Chromium (~100 MB) for rendering. Then:

npx skills add openfilm/openfilm

After that, you literally prompt your agent the way you'd brief a human editor: "Make a 20-second launch video for my CLI, with music." It builds the film, and Studio opens beside it with a chat panel — generate, then refine, in the same loop.

Prefer not to touch the terminal? Grab the desktop app for macOS/Windows from openfilm.dev — Studio with chat built in, ffmpeg and the browser bundled. You can also start from finished examples instead of a blank page:

npx -y openfilm@latest open https://github.com/openfilm/examples/tree/main/one-prompt

The cost model — read this part

Making, editing, and rendering films: free, no account needed. The catch is where the AI media comes from: AI-generated voice, music, images, and video connect through your own API keys (ElevenLabs, OpenAI, Google, fal) in Settings → Providers. You pay the providers, not OpenFilm. Budget accordingly — the editor is free, the generative ingredients are not.

Who this is for — and who it's not

For: explainer videos, launch videos, animated explainers, faceless knowledge content — anything where the "footage" can be motion graphics plus real clips.
Not for: replacing photorealistic AI generation pipelines (Veo, Sora, LTX). Think of OpenFilm as the editor + motion-graphics agent layer that sits on top of whatever footage you have.

Verdict. The editability angle is genuinely new — AI video that isn't a one-shot lottery ticket. It's a young project (launched October 2026), so expect rough edges. But the architecture — plain HTML plus a real timeline — means every rough edge is inspectable and fixable. That's a bet worth watching.

Repo: github.com/openfilm/openfilm · 975 stars and climbing · open source · updated daily. Information current as of October 2026; verify install steps against the repo README before running.