reel-cutter

Auto-cut talking-head videos. Removes mistakes, repeated phrases, filler words ("eh", "vale", "pues"), self-corrections, and long silences. Output is the original video minus the bad takes.

Pipeline: denoise → transcribe (WhisperX) → clean script (Claude) → cut (ffmpeg) → verify.

Install

# Recommended: install as a uv tool (isolated, on your $PATH)
uv tool install git+https://github.com/Racks-Labs/reel_cutter.git

# Or with pipx
pipx install git+https://github.com/Racks-Labs/reel_cutter.git

# Or with plain pip
pip install git+https://github.com/Racks-Labs/reel_cutter.git

Requirements:

• Python 3.10–3.12
• ffmpeg available on $PATH (brew install ffmpeg / apt install ffmpeg)
• An Anthropic API key

First-time setup

reel-cutter setup
# Prompts for API key, saves to ~/.config/reel-cutter/.env (mode 600)

Usage

reel-cutter run my_video.mp4

Output appears in my_video_reel_edit/my_video_edited.mp4 next to the source video.

Show the full manual:

reel-cutter help

Check what's loaded:

reel-cutter env-status

How configuration is found

.env files are loaded in this order (later overrides earlier):

1. ~/.config/reel-cutter/.env — global user config (written by setup)
2. ./reel-cutter.env — project-local, explicit
3. ./.env — project-local, conventional
4. Shell environment variables — always wins

Required key: ANTHROPIC_API_KEY.

Tuning

# English
reel-cutter -l en run video.mp4

# Faster (smaller Whisper model)
reel-cutter -m base run video.mp4

# Single-shot (no auto-tuning loop)
reel-cutter run video.mp4 --no-tune

# Force re-run all steps
reel-cutter run video.mp4 -f

See reel-cutter help for every flag.

Step-by-step (advanced)

reel-cutter denoise video.mp4
reel-cutter transcribe video.mp4
reel-cutter clean video.mp4
reel-cutter cut video.mp4
reel-cutter verify video.mp4

Each step caches its output as JSON next to the video, so you can iterate on a single step without re-running the whole pipeline.

License

MIT