video-use enables chat-driven editing
- browser-use’s open-source `video-use` project gained traction this week as a chat-driven video editor for Claude Code, promising to turn a folder of raw clips into `edit/final.mp4` with conversational prompts. - The GitHub repo showed about 4,000 stars and 566 forks on April 26, with features including word-level transcript cuts, auto color grading, subtitle burn-ins, and animation overlays generated in parallel. - The project joins a growing wave of prompt-first editing tools that replace timeline interfaces with agent workflows and external editors. (buttercut.io)
Video editing is usually a timeline problem: drag clips, trim pauses, fix color, add captions, export. `video-use` turns that into a chat prompt inside Claude Code. (github.com) The project comes from the browser-use team and describes itself as “100% open source.” Its setup is simple on paper: drop raw footage into a folder, open Claude Code there, and ask it to “edit these into a launch video.” (github.com) On April 26, the GitHub repo showed about 4,000 stars and 566 forks. The README says the output lands as `edit/final.mp4` next to the source files rather than inside a traditional non-linear editor timeline. (github.com 1) (github.com 2) The underlying trick is that the language model does not “watch” the footage like a human editor. The README says it reads a word-level transcript from ElevenLabs Scribe, plus selected visual snapshots, to make cut decisions with timestamp precision. (github.com) That matters because most of the early automation is built around spoken video: interviews, tutorials, talking heads, travel clips. `video-use` says it can remove filler words, trim dead space, apply 30-millisecond audio fades, burn subtitles, and run custom color-grading chains through FFmpeg. (github.com) The project also stores session context in a `project.md` file so a later editing session can pick up where the last one stopped. Its install guide says it can be wired into Claude Code, Codex, Hermes, or OpenClaw, as long as the agent can discover the skill files. (github.com 1) (github.com 2) A second layer handles graphics. The repo includes a `manim-video` skill for generating animated explainers, diagrams, and overlays, and the main README says Manim, Remotion, or Python Imaging Library jobs can run in parallel sub-agents. (github.com 1) (github.com 2) The catch is that this is not a one-click consumer app. The install docs require FFmpeg, Python dependencies, and an ElevenLabs API key for transcription, and the workflow still depends on the user approving strategy choices inside an agent session. (github.com) It is also not alone. ButterCut, another open-source Claude Code video workflow, analyzes footage and produces timelines for Final Cut Pro, Adobe Premiere Pro, and DaVinci Resolve instead of exporting a finished MP4 directly. (buttercut.io) That leaves `video-use` in a specific lane: chat-first editing for creators who would rather describe an edit than build it shot by shot. The repo’s pitch is blunt — raw footage in, conversation in the middle, final MP4 out the other end. (github.com)
