Your photos stay home; the machine learning moves in

What it does

LibrePhotos is a self-hosted photo and video archive that scans your filesystem, extracts EXIF data, and builds a browsable timeline. It runs face recognition, scene detection, and semantic search locally using models like places365 and im2txt, then serves everything through a React frontend with MapLibre maps. Multi-user support means your whole household can share the same instance without sharing accounts.

The interesting bit

The project recently swallowed five separate repositories—backend, frontend, mobile app, docs, and deployment configs—into one monorepo while preserving the full commit history. That administrative feat is almost as impressive as the ML pipeline, and it suggests the maintainers are serious about long-term hygiene rather than just duct-taping containers together.

Key highlights

• Supports RAW photos and videos, not just JPEGs, with ImageMagick and FFmpeg handling conversions
• Generates event-based albums like “Thursday in Berlin” from reverse geocoding and timestamps
• ML stack includes face_recognition, hdbscan clustering, and semantic captioning, but the README warns you’ll want 8 GB+ RAM for smooth operation
• Ships with a React Native Android client, Django 5 API, and interactive Swagger/ReDoc documentation
• Runs via Docker Compose with Nginx, PostgreSQL, and a Django-Q2 task queue

Caveats

• The ML features are explicitly memory-intensive; 8 GB+ RAM and an SSD are strongly recommended
• Mobile support is currently limited to an Android client built with React Native
• The README lists ML capabilities like semantic search and image captioning, but offers no detail on model accuracy or indexing speed

Verdict

Self-hosting enthusiasts with a spare server and a large photo library will find LibrePhotos a capable alternative to cloud lock-in. If your hardware is below the 4 GB minimum or you need iOS support, this isn’t your tool yet.