browser-use's video-use Lets AI Edit Raw Footage With Plain English
video-use lets coding agents edit raw footage into a polished final.mp4 using transcripts, ffmpeg, and a self-correcting render loop — no timeline editor required
Takeaways
- video-use is a new open-source tool from the browser-use team that lets coding agents edit raw video footage via natural language commands, outputting a polished
final.mp4. - The LLM never processes raw frames; instead it reads a compact ~12KB word-level transcript from ElevenLabs Scribe, slashing token usage from a theoretical 45M tokens to a handful of PNGs.
- The pipeline covers filler-word removal, color grading, 30ms audio fades, burned-in subtitles, and animated overlays via Manim, Remotion, or HyperFrames — all in parallel sub-agents.
- A built-in self-evaluation loop checks every cut boundary for visual jumps and audio pops before surfacing the preview, with up to 3 auto-fix retries.
- Works with Claude Code, Codex, Hermes, and any shell-access agent; requires Python 3.10+, ffmpeg, and an ElevenLabs API key for transcription.
- The repo is MIT-licensed, fully open source, and crossed 9,000 GitHub stars shortly after launch.
The team behind browser-use , the project that taught AI agents to navigate the web by reading structured HTML instead of screenshots , just applied the same philosophy to video editing. Their new open-source tool, video-use, lets you drop raw footage into a folder, type a plain-English instruction to a coding agent, and get back a final.mp4 with cuts, color grading, subtitles, and animated overlays. No timeline. No drag-and-drop. No Premiere.
The repo crossed 9,000 GitHub stars within days of launch and has already accumulated over 1,100 forks , a signal that this hit a real nerve. The pitch is simple: traditional video editing requires specialized software like Premiere, DaVinci Resolve, or Final Cut and months of practice. video-use flips the model: the LLM becomes your editor, and you become the creative director.