video-use: edit your raw phone footage by chatting with Claude Code

What it does and who it is for

video-use, from the Browser Use team, edits real footage by conversation. You put your raw takes in a folder, open Claude Code or Codex there, and say what you want. It transcribes the audio, reads the transcript with word-level timings, proposes a cut, waits for your OK, then renders final.mp4.

The trick is that the AI reads your video instead of watching every frame. The transcript is the main view; it only pulls picture strips at the points where it needs to decide something, like which of two takes is cleaner. Per the README, it can remove filler words and dead air, colour-grade each segment, fade audio at every cut so you do not hear a pop, burn in subtitles, add animated overlays (it can use HyperFrames for those), and check its own render at each cut before showing you. It keeps notes in a project.md so next week's session remembers the project.

It suits owners and professionals who record themselves on a phone (a product demo, a how-to, a talk to camera) and do not want to learn an editor. It is installed on my machine as a Claude Code skill, set up the manual way below.

What you need first

• Claude Code or Codex.
• Python with uv (or pip), and FFmpeg. On a Mac: brew install ffmpeg uv.
• An ElevenLabs API key. video-use uses ElevenLabs' Scribe service for transcription. ElevenLabs is a paid service with its own plans; check their pricing before you run a lot of footage.
• About 20 minutes to install.
• The skill is free, but it runs inside a coding agent such as Claude Code or Codex, which may need a paid plan.

Install (checked 30 Sep 2026)

The README offers a setup prompt you paste into your agent, which then reads the repo's install.md and does the steps. If you prefer to do it by hand, these are the README's manual commands as of 30 Sep 2026:

git clone https://github.com/browser-use/video-use ~/Developer/video-use
ln -sfn ~/Developer/video-use ~/.claude/skills/video-use
cd ~/Developer/video-use
uv sync
brew install ffmpeg
cp .env.example .env

For Codex, link it into ~/.codex/skills/video-use instead. pip install -e . works in place of uv sync. brew install yt-dlp is optional, for pulling online sources.

Open .env in a text editor and put your ElevenLabs key after ELEVENLABS_API_KEY=. Type it into the file yourself; do not paste keys into a chat.

I did not run a fresh install for this guide: it needs a real API key and sends audio to ElevenLabs. The copy on my machine uses exactly this layout (a clone in ~/Developer/video-use, linked into the Claude skills folder).

First use

Two approvals come at different moments. Before the first run, decide that these clips may go to ElevenLabs for transcription and that you accept the cost for their length. The edit plan comes later, after transcription.

Put three or four raw clips in a folder, then:

cd ~/Videos/sunday-promo
claude

In the session:

What the README describes (I did not run it for this guide): it lists the clips it found and how long each is, transcribes them, and comes back with a plain-English plan (which takes, what order, what style of captions). Nothing is cut until you say OK. The result lands in an edit/ folder next to your clips, as edit/final.mp4. Watch it all the way through, then ask for changes in words: "tighten the intro", "captions smaller".

How it chains with other entries

• HyperFrames makes designed video from scratch; video-use can call it for animated titles and overlays on top of real footage.
• Check the AI's answer before a customer sees it is the same habit for the last step: watch the final file before anyone else does.

Watch-outs

• Your footage leaves your machine in two ways. The audio goes to ElevenLabs for transcription. The transcript, and the picture strips it pulls at decision points, go into your coding agent's AI model as part of the chat. Do not feed it client calls, customer faces or private conversations unless the people in them agreed.
• Cost. ElevenLabs usage is billed by them, not by video-use.
• Posting is yours. video-use produces a file. Watching it and deciding to post it stays with a person.
• It follows your words. If the plan it proposes is wrong, say so before you approve; fixing the plan is cheaper than re-rendering.

Link and credit

Repo: browser-use/video-use, MIT licence, by Browser Use. The editing method, rules and helpers are theirs; this page is my own walkthrough.