video-use

video-use is a 100% open-source project from the Browser Use team that lets you edit videos with an AI coding agent. You drop raw footage into a folder, chat with an agent such as Claude Code, Codex, or Hermes, and get a final.mp4 back. It works for any content: talking heads, montages, tutorials, travel footage, and interviews, without presets or menus.

The project is hosted on GitHub at github.com/browser-use/video-use, where it has gathered over 18,700 stars and 2,343 forks since its release in April 2026. It is MIT-licensed and designed to plug into any agent with shell access, including Claude Code, Codex, Hermes, and OpenClaw.

The core idea mirrors Browser Use's approach to web automation: instead of dumping 30,000 video frames into the model (about 45 million tokens of noise), video-use gives the agent a compact transcript and a few on-demand visual composites, so it can make precise, word-boundary-level editing decisions with a tiny token budget.

• Cuts out filler words (umm, uh, false starts) and dead space between takes
• Auto color grades every segment: warm cinematic, neutral punch, or any custom ffmpeg chain
• 30ms audio fades at every cut so you never hear a pop
• Burns subtitles in your style: 2-word UPPERCASE chunks by default, fully customizable
• Generates animation overlays via HyperFrames, Remotion, Manim, or PIL, spawned in parallel sub-agents, one per animation
• Self-evaluates the rendered output at every cut boundary before showing you anything
• Persists session memory in project.md so next week's session picks up where you left off

The LLM never watches the video. It reads it, through two layers that together give it everything needed to cut with word-boundary precision.

Layer 1: Audio Transcript (Always Loaded)

One ElevenLabs Scribe call per source gives word-level timestamps, speaker diarization, and audio events such as laughter, applause, or sighs. All takes pack into a single ~12KB takes_packed.md file, which becomes the LLM's primary reading view:

## C0103  (duration: 43.0s, 8 phrases)
  [002.52-005.36] S0 Ninety percent of what a web agent does is completely wasted.
  [006.08-006.74] S0 We fixed this.

Layer 2: Visual Composite (On Demand)

The timeline_view tool produces a filmstrip plus waveform plus word-labels PNG for any time range. It is called only at decision points: ambiguous pauses, retake comparisons, and cut-point sanity checks. Where a naive approach would feed the model 30,000 frames times 1,500 tokens, video-use works from about 12KB of text plus a handful of PNGs. It is the same idea as browser-use giving an LLM a structured DOM instead of a screenshot, but for video.

Transcribe ──> Pack ──> LLM Reasons ──> EDL ──> Render ──> Self-Eval
                                                        │
                                                        └─ issue? fix + re-render (max 3)

The self-eval loop runs timeline_view on the rendered output at every cut boundary, catching visual jumps, audio pops, and hidden subtitles. You only see the preview after it passes. If an issue is found, the agent fixes and re-renders, up to three attempts.

1. Text + on-demand visuals. No frame-dumping. The transcript is the surface.
2. Audio is primary, visuals follow. Cuts come from speech boundaries and silence gaps.
3. Ask, confirm, execute, self-eval, persist. Never touch the cut without strategy approval.
4. Zero assumptions about content type. Look, ask, then edit.
5. 12 hard rules, artistic freedom elsewhere. Production-correctness is non-negotiable. Taste is not.

The full production rules and editing craft live in the project's SKILL.md file, which the agent reads for daily usage.

One-Line Agent Setup

Paste this prompt into Claude Code, Codex, Hermes, OpenClaw, or any agent with shell access, and the agent handles the rest:

Set up https://github.com/browser-use/video-use for me.

Read install.md first to install this repo, wire up ffmpeg, register the skill with whichever agent you are running under, and set up the ElevenLabs API key - ask me to paste it when you need it. Then read SKILL.md for daily usage, and always read helpers/ because that is where the editing scripts live. After install, do not transcribe anything on your own - just tell me it is ready and wait for me to drop footage into a folder.

Manual Install

# 1. Clone and symlink into your agent's skills directory
git clone https://github.com/browser-use/video-use ~/Developer/video-use
ln -sfn ~/Developer/video-use ~/.claude/skills/video-use        # Claude Code
# ln -sfn ~/Developer/video-use ~/.codex/skills/video-use       # Codex

# 2. Install deps
cd ~/Developer/video-use
uv sync                         # or: pip install -e .
brew install ffmpeg             # required
brew install yt-dlp             # optional, for downloading online sources

# 3. Add your ElevenLabs API key
cp .env.example .env
$EDITOR .env                    # ELEVENLABS_API_KEY=...

ffmpeg is required for rendering and processing. yt-dlp is optional and only needed to download online sources. An ElevenLabs API key powers the Scribe transcription layer (word-level timestamps, speaker diarization, and audio events).

Point your agent at a folder of raw takes and start a session:

cd /path/to/your/videos
claude    # or codex, hermes, etc.

Then in the session, type something like: "edit these into a launch video". The agent inventories the sources, proposes a strategy, waits for your OK, then produces edit/final.mp4 next to your sources. All outputs live in <videos_dir>/edit/, so the skill directory stays clean. Session memory persists in project.md, so a session next week picks up where the last one left off.

For always-on editing from your own VPS or Telegram, the project recommends running the agent through Browser Use Box.

video-use occupies a new niche between traditional NLEs, AI video generators, and auto-editing apps.

Criteria	video-use	Traditional NLE (Premiere)	AI Video Generators	Auto-Editing Apps
Input	Existing raw footage	Existing raw footage	Text prompt	Existing raw footage
Control	Natural language, word-level	Full manual timeline	Prompt-level	Template-level
Filler removal	Automatic, word-boundary	Manual scrubbing	Not applicable	Sometimes
Self-evaluation	Automated at every cut	Manual review	None	Limited
Open source	Yes, MIT	Proprietary, paid	Proprietary services	Mixed
Learning curve	Chat in plain language	Steep timeline mastery	Prompt craft	Minimal

video-use vs Traditional NLEs: editors like Premiere Pro or DaVinci Resolve give frame-perfect manual control, which professional editors need for complex projects. video-use excels at the high-volume, low-ceremony jobs: cutting talking-head footage, removing fillers, and assembling montages, all driven by conversation instead of timeline mastery.

video-use vs AI Video Generators: text-to-video models create footage from prompts but cannot clean up footage you already recorded. video-use does the opposite: it refines real footage with precise, transcript-aware cuts, color grading, subtitles, and overlays, so the two approaches complement rather than replace each other.

video-use vs Auto-Editing Apps: consumer auto-editors apply fixed templates with limited control. video-use is open, scriptable, and style-driven: the editing rules live in SKILL.md, the subtitle style is customizable, and any ffmpeg chain can serve as a grade, so the output can match a specific brand or creator style.

video-use is part of the Browser Use ecosystem, the open-source project that gives AI agents a structured view of the web. The same design philosophy, structured text plus on-demand visuals, is applied here to video. It integrates with Browser Use Cloud and Browser Use Box for hosted and always-on runs.

The animation overlay layer plugs into HyperFrames, Remotion, Manim, or PIL, spawned in parallel sub-agents with one animation per sub-agent. Transcription is powered by ElevenLabs Scribe, and rendering runs on ffmpeg.

With over 18,700 stars and 2,343 forks, video-use is one of the fastest-growing open-source AI video projects on GitHub, MIT-licensed for free commercial and personal use.