How to use video-use: edit video with Claude Code

What you need before installing video-use

video-use runs inside an agent that has shell access, such as Claude Code or Codex. It also needs ffmpeg, uv or pip for Python dependencies, and an ElevenLabs API key, because transcription uses the ElevenLabs Scribe service.

• Claude Code, Codex, or another agent with shell access
• ffmpeg installed on your machine
• uv or pip for Python dependencies
• An ElevenLabs API key

Install video-use with the manual steps

The README gives a manual install that clones the repo, links it into your agent's skills directory, installs dependencies and adds your API key. The first command is for Claude Code; a Codex path is shown as a commented alternative.

The README lists brew install ffmpeg as required and brew install yt-dlp as optional, for downloading online sources.

• Clone the repo and symlink it into the skills directory for your agent
• Run uv sync, or pip install -e . as the alternative
• Install ffmpeg (required) and yt-dlp (optional)
• Copy .env.example to .env and set ELEVENLABS_API_KEY

Let the agent do the setup instead

The README also offers a setup prompt you can paste into Claude Code, Codex or another agent. The agent is told to read install.md, wire up ffmpeg, register the skill and ask you to paste your ElevenLabs key when it needs it.

The prompt tells the agent not to transcribe anything after install. It should report that it is ready and wait for you to add footage.

Run your first edit session

Put your raw takes in one folder, change into it and start your agent there. Then describe the edit in plain language.

video-use inventories the sources and proposes a strategy, then waits for your approval before it touches any cut. All outputs go into an edit/ folder inside your videos directory, so the skill directory stays clean.

• Open a terminal in the folder with your footage
• Start Claude Code, Codex or another agent
• Ask for the edit, for example: edit these into a launch video
• Approve the strategy the agent proposes
• Find the result at edit/final.mp4 next to your sources

What video-use does to your footage

According to the README, the agent cuts filler words such as umm, uh and false starts, and removes dead space between takes. It auto color grades each segment, adds 30ms audio fades at every cut, and burns in subtitles, by default as 2-word uppercase chunks.

The README says the model reads a text transcript rather than watching every frame. One ElevenLabs Scribe call per source gives word-level timestamps and speaker labels, packed into a single transcript file. The agent looks at visual composites only at decision points, and it checks the rendered output at each cut boundary, up to three fix-and-re-render attempts.

• Filler word and dead space removal
• Color grading, with presets such as warm cinematic or neutral punch, or a custom ffmpeg chain
• 30ms audio fades at cuts
• Subtitles burned in, customizable
• Animation overlays via HyperFrames, Remotion, Manim or PIL
• Session memory in project.md

Keep context between sessions

video-use persists session memory in project.md, so a later session can pick up earlier context. This is useful when you return to an edit a week later and want the agent to remember decisions you already made.