Immich: The Self-Hosted Google Photos That Actually Works

Google Photos puts your entire photo library on Google's servers to feed the same ad ecosystem as everything else. Immich is the exit: automatic backup to your own server, facial recognition, semantic search, albums, sharing. 106,000+ GitHub stars, AGPL-3.0, v3.0.0 (July 2026). The most popular self-hosted photo manager by a large margin.

What Immich is

Immich (immich.app, github.com/immich-app/immich) is a high-performance self-hosted photo and video platform. NestJS backend, SvelteKit web, Flutter iOS/Android apps, PostgreSQL with VectorChord, Redis. Docker Compose deployment. AGPL-3.0.

Core features

Automatic backup - open the app, photos back up in the background. Duplicate detection, selective album backup, iOS and Android. Free Up Space (v2.5) removes local copies of photos already on your server.

Facial recognition - clusters faces across your library by person. Name the cluster, all photos of that person are tagged. Runs locally, no data sent externally.

CLIP semantic search - search by content ("beach sunset") using CLIP embeddings. Metadata, EXIF, location, date search. OCR (v3.0): text in photos is extracted and searchable.

Albums and sharing - shared albums, partner sharing, public share links, multi-user upload. Stacked photos for burst shots.

Global map - GPS-tagged photos on a world map. Browse by location.

Memories - "On this day" across past years, same as Google Photos.

Non-destructive editing (v2.5 web, v3.0 mobile) - crop, rotate, adjust without touching the original file. Edits in the database, originals untouched, revert any time. Web and mobile share the same system.

Workflows (v3.0, preview) - automate library actions via drag-and-drop builder. Triggers, filters, actions. Automation foundation with more coming.

Video - upload, playback, HLS real-time transcoding (v3.0). Plays in browser without offline processing.

RAW, HEIC, Live Photos, Motion Photos - all supported with thumbnail generation and playback.

Multi-user - admin panel, storage quotas, OAuth/OIDC, API keys, granular permissions.

Storage templates - user-defined file structure on disk. Predictable hierarchy, not opaque blob storage.

Machine learning

Dedicated immich-machine-learning container for face detection/recognition, CLIP, and OCR. CPU by default. GPU: CUDA, OpenVINO, ONNX. Can point at a remote ML host to offload inference from the main server.

Self-hosting

Docker Compose: immich-server, immich-machine-learning, PostgreSQL, Redis. Minimum 4 GB RAM, 8 GB+ for ML. Any x86_64 or ARM64. Install takes 10 minutes. Update by changing IMMICH_VERSION=v3 in .env and pulling new images. Database migrations run automatically.

vs Google Photos

Google Photos: zero setup, tight Android integration, Pixel AI. Immich: your data on your hardware, no ads, no ML training on your photos, unlimited storage bounded by disk, open source, no subscription risk.

Who it is for

Good fit: NAS or home server users; families wanting a private shared library; photographers with large RAW collections; teams integrating with Dawarich, Photoprism, or other self-hosted tools via the Immich API.

Not ideal: no server and no interest in running one; Google Lens AR features needed; organizations requiring formal SLA.

My take

Immich is the answer to "what instead of Google Photos" and the 106k stars are not hype. Mobile backup is reliable, ML search is good enough to replace manual organization, and v3.0 makes it feel complete. One thing to get right before deploying: backups. Immich stores your originals. Losing the server means losing photos. 3-2-1 backup before you migrate.