video-use
A typed Node.js CLI that lets an AI agent edit raw video footage conversationally, transcribing clips, reasoning over an edit-decision list, and rendering the result with ffmpeg.
video-use gives an agent phrase-level, word-level transcripts with speaker diarization from ElevenLabs Scribe, packed into a silence-aware Markdown format meant to be token-efficient, instead of raw video frames. The agent reads the packed transcript, proposes a cut strategy, writes an edit-decision list as JSON, and the CLI renders the final video with ffmpeg.
Its render pipeline applies a per-segment grade, short audio fades, HDR tone-mapping, overlay compositing and subtitle burn-in last, and generates filmstrip-and-waveform composite images to help drill into specific cut points. An optional Redis cache persists transcripts across sessions when enabled.
It requires Node.js 18 or newer, ffmpeg and ffprobe, and an ElevenLabs API key for transcription; it is written in strict TypeScript with a Vitest test suite and an ESLint configuration, and is MIT licensed.