Extract representative frames from thousands of home videos and describe them with a local AI.
Make 1,292 home videos searchable without uploading them anywhere.
PySceneDetect (AdaptiveDetector) plus FFmpeg split scenes; Microsoft Florence-2 Base runs locally to caption each scene; results in PostgreSQL media_db. No cloud APIs, no pay-per-token.
Pipeline defined (5 years of footage); no cloud cost (0 USD declared); implementation ongoing.
Full architectural rules to be published by the author.