LibrePhotos — Self-Hosted AI-Powered Photo Management

Introduction

LibrePhotos is a self-hosted photo management platform that uses machine learning for automatic face recognition, scene classification, and object detection. It provides a Google Photos-like experience while keeping all your images and metadata on your own server, giving you full ownership of your photo library.

What LibrePhotos Does

• Automatically detects and clusters faces across your photo library for people albums
• Classifies scenes and objects using pre-trained models for smart search
• Extracts and displays EXIF metadata including GPS coordinates on an interactive map
• Generates timeline views with automatic grouping by date and location
• Supports multi-user setups with individual libraries and sharing capabilities

Architecture Overview

LibrePhotos uses a Django backend with Celery for asynchronous task processing. Machine learning inference runs on-device using PyTorch models for face detection, recognition, and image classification. PostgreSQL stores metadata and user data. The frontend is a React single-page application. Photos are stored on the filesystem and indexed by a background scanner that extracts metadata, generates thumbnails, and runs ML classification pipelines.

Self-Hosting & Configuration

• Deploy using the official librephotos-docker repository with Docker Compose
• Mount your existing photo directories as volumes in the container configuration
• Set ADMIN_EMAIL and ADMIN_PASSWORD in .env for the initial admin account
• Configure the scan directory path where LibrePhotos will discover new photos
• Optionally enable GPU passthrough in Docker for faster ML inference on NVIDIA hardware

Key Features

• Face detection and recognition with automatic clustering into people albums
• Scene and object classification for intelligent photo search
• Interactive map view using GPS data from photo EXIF metadata
• Background photo scanning with automatic thumbnail generation
• Multi-user support with per-user libraries and sharing

Comparison with Similar Tools

• PhotoPrism — Similar AI features but uses TensorFlow; LibrePhotos uses PyTorch and offers face clustering
• Immich — Mobile-first with companion app; LibrePhotos focuses on server-side ML processing
• Photoview — Lightweight gallery without ML features; LibrePhotos adds face and object recognition
• Nextcloud Photos — Part of a larger ecosystem; LibrePhotos is purpose-built for photo management
• Piwigo — Traditional gallery manager; LibrePhotos provides modern AI-driven organization

FAQ

Q: How much storage overhead does LibrePhotos add? A: Thumbnails and ML embeddings typically add 5-10% on top of your original photo storage.

Q: Can I use LibrePhotos without a GPU? A: Yes, ML inference works on CPU. A GPU speeds up the initial scan but is not required for daily use.

Q: Does it support RAW photo formats? A: LibrePhotos supports common RAW formats through its image processing pipeline and generates viewable thumbnails.

Q: How does face recognition training work? A: You manually confirm or correct a few face clusters, and LibrePhotos refines its recognition model for your library.

Sources

• https://github.com/LibrePhotos/librephotos
• https://docs.librephotos.com