moh4696/freecut

freecut is a fork of an open-source project called video-use that lets you edit video files by chatting with an AI coding assistant like Claude Code or Codex. The original project required a paid ElevenLabs speech transcription API key. freecut removes that requirement by replacing the transcription layer with free alternatives that run locally on your machine. The editing process works like this: you point the tool at a folder of raw video clips, open your AI assistant in that folder, and describe what you want. The assistant reads the transcripts, builds an edit list, waits for your approval, then calls ffmpeg to produce a final video file called final.mp4. It never watches the actual video frames directly. Instead it reads a compact text transcript of every word spoken, with timestamps, so it can make precise cuts based on speech boundaries without processing thousands of video frames. The default transcription backend is Whisper, a free, open-source speech-to-text model from OpenAI that runs entirely on your own computer with no API key needed. For videos with multiple speakers, an optional backend called VibeVoice-ASR can identify who is talking when, but this requires a machine with an NVIDIA GPU and is not available on most Macs. The original ElevenLabs backend is still supported if you already have a key. Beyond cutting, freecut can remove filler words and dead air, apply color grading, add subtitles, and burn in animation overlays. After each render it checks its own output at every cut point before showing you the result, and it saves notes about your project so a later session can pick up where you left off. Installation takes a few minutes: clone the repo, install Python dependencies with uv or pip, install a Whisper package, and ensure ffmpeg is on your system. No API keys are required for the default local-Whisper setup.