Self-hosted, white-label AI clipping platform for agencies, clipping communities and streamer managers: long videos and streams become scored, face-tracked, word-captioned vertical clips on the buyer's own server, and brands run paid clipping campaigns with a review queue and a payout ledger.

What it does

• Sources: resumable upload, library, 720p proxy, waveform and thumbnails
• Import from a link through the operator's own yt-dlp (off by default)
• Transcription with word timings: ElevenLabs Scribe, Whisper-compatible HTTP, local whisper.cpp
• Moment picking by Claude or a no-key heuristic, snapped to sentence boundaries
• Score from 7 named signals with a reason each
• Complete-thought check with one-click fix; titles, hooks, hashtags; workspace preferences learned from approvals
• Face-tracked reframing (YuNet ONNX), speaker choice by mouth movement, layouts fill/fit/split/stack/screen
• 18 word-by-word caption presets drawn by one canvas module for browser preview and server render

You need

• Node.js 22.12 or newer
• ffmpeg and ffprobe on the server (installed by the buyer, or by the Dockerfile from Debian packages; not bundled)
• A VPS (no shared hosting): CPU video rendering and face detection run on the server. Sizing unverified; docs/KNOWHOW.md
• Domain + https reverse proxy
• ADMIN_SECRET_KEY (32 random bytes hex) to seal saved API keys