# video-use

Agent skill that turns a folder of raw takes into a finished MP4 from a chat prompt.

A Claude Code, Codex, or Hermes skill that reads footage through an ElevenLabs word-level transcript instead of frames, proposes a cut, renders it with ffmpeg, and self-checks the output at every cut boundary. Filler-word removal, color grades, burned subtitles, and animation overlays with HyperFrames, Remotion, or Manim are the built-in jobs.

Alternative to: Descript, CapCut

- Capabilities: adaptable, operable
- Vertical: video-streaming
- Shape: skill
- Form factor: skill
- Execution model: hybrid
- Provenance: unknown (recorded, never ranked)
- Repo: https://github.com/browser-use/video-use (MIT), 23605 stars, last commit 2026-08-30
- Last activity: 2026-08-30
- Page: apps/browser-use--video-use.html
- JSON: apps/browser-use--video-use.json

## Use it when

- I have a folder of talking-head takes and want a clean cut with subtitles without opening an editor
- I want my coding agent to edit tutorials and launch videos the same way it edits code

## Gotchas

- ElevenLabs Scribe is the only transcription path; no key, no cuts
- ffmpeg must be on PATH; yt-dlp only for online sources
- The agent proposes a strategy and waits for approval before cutting, so unattended runs need an agent configured to auto-approve

## Editions

- Free: Free, open source
  - source: https://github.com/browser-use/video-use
  - installable: git clone https://github.com/browser-use/video-use ~/Developer/video-use && ln -sfn ~/Developer/video-use ~/.claude/skills/video-use
- Browser Use Cloud: Free, closed source
  - hosted: https://cloud.browser-use.com/v4

## Adaptable: make it yours

Effort: config-only (6 things to edit)

Set ELEVENLABS_API_KEY in .env, then change editing behavior by editing SKILL.md (the 12 production rules and default subtitle style) or the ffmpeg chains in helpers/. No code change is needed for a different look; the agent reads the prompt files each session.

- Env vars: ELEVENLABS_API_KEY
- Config files: .env, pyproject.toml
- Prompt files: SKILL.md, install.md, helpers/
- Accounts you'll need: ElevenLabs

## Operable: what an agent can drive

Operability: native. Interface: cli at https://github.com/browser-use/video-use/blob/main/SKILL.md (published by the builder, local).

### Reach (complete coverage)

- Cut filler words and dead space between takes: reachable. README 'What it does'; cuts come from the Scribe transcript and silence gaps, executed by helpers/ scripts
- Auto color grade every segment: reachable. README: warm cinematic, neutral punch, or any custom ffmpeg chain
- Audio fades at every cut: reachable. README: 30ms audio fades at every cut
- Burn styled subtitles: reachable. README: 2-word UPPERCASE chunks by default, fully customizable
- Generate animation overlays: reachable. README: with HyperFrames, Remotion, Manim, or PIL in parallel sub-agents
- Self-evaluate the rendered output: reachable. README: timeline_view run on rendered output at every cut boundary, max 3 fix loops
- Persist session memory across sessions: reachable. README: project.md next to the sources
- Download online sources: reachable. README manual install: yt-dlp optional for downloading online sources

## Code stack

- Languages: Python
- Frameworks: ffmpeg
- Package manager: uv
- Database: none
- Hosting target: local

## Model stack

- Providers: ElevenLabs, any coding agent with shell access
- Models: ElevenLabs Scribe
- Keys you need (names only): ELEVENLABS_API_KEY
- Prompts live in: SKILL.md, install.md
- Tools: helpers/ editing scripts, timeline_view

## Skills

pure: Repo is a skill: root SKILL.md plus a manim-video sub-skill; helpers/ holds Python helper scripts the skill calls, not a CLI or app.
- Paths: SKILL.md, skills/manim-video/SKILL.md

## Evidence and confidence

- formFactor: README: symlinked into ~/.claude/skills or ~/.codex/skills; SKILL.md marker in shortlist (confidence 0.95)
- executionModel: README pipeline: Transcribe -> Pack -> LLM Reasons -> EDL -> Render -> Self-Eval; fixed stages, model decides cuts (confidence 0.85)
- operability: README: setup is a prompt pasted into Claude Code, Codex, Hermes, or Openclaw; no GUI in the tree; the pipeline stage LLM Reasons has no non-model substitute (confidence 0.95)
- effort: .env.example contains ELEVENLABS_API_KEY only; SKILL.md holds the editing rules (confidence 0.85)
- editions: MIT LICENSE; README links Browser Use Cloud with no price shown (confidence 0.7)
- active: GitHub pushedAt 2026-08-30
- reach:  (confidence 0.85)

## Open questions

- Browser Use Cloud pricing for running video-use hosted was not found on the README; check cloud.browser-use.com
- Provenance: browser-use is an AI-agent company; no statement on whether the skill itself was vibe-coded

## Sources

- https://github.com/browser-use/video-use (sourcing:video-streaming:community)