FrameWright

FrameWright is a local-first video pipeline that drives post-production editing from an editor-centric approach. By simply placing a video in a recording folder, you can consistently perform transcription (whisper.cpp), silence detection (ffmpeg), generation of semantic cut proposals and chaptering via LLM, human approval, and final rendering. The main flow is: open the project in the editor → automatically execute transcription and silence detection as needed → LLM proposes a cut plan (plan) → editor adjusts via GUI or direct JSON editing → confirm via preview → pass the approval gate → Full HD output via ffmpeg and a proprietary synthesis engine (WebCodecs + WebGPU). Notable features: - Completely local processing: Transcription and silence detection operate offline. Only transcription text is passed to the LLM; video and audio are never sent externally. - Support for various canvases: 16:9, 9:16, 1:1, etc. The derivative project feature allows for different output sizes while sharing the original video. - Extended canvas: An advanced workflow that records screen and camera into one file and synthesizes a bottom-right picture-in-picture wipe. - Automatic audio processing: Automatic mixing of microphone and system audio, loudness normalization to -14 LUFS, and automatic BGM ducking and looping. - Human approval gate: AI-driven automatic editing cannot proceed to final rendering unless approved; by design, the AI cannot act as the approver. - Agent integration: Integrated into Claude Desktop and others as an MCP server to provide safe editing tools (approval and rendering are not permitted). Setup is completed by installing macOS, Node.js 23.6+, Homebrew, ffmpeg, and whisper-cpp, followed by npm install. The doctor command allows for environment checks, and it can be tried immediately with a sample project. FrameWright is an opinionated tool specialized in a single workflow from recording screen demos and commentary to YouTube publication. It is not intended as a replacement for general-purpose NLEs, but rather aims to automate primary editing and improve finishing efficiency.