Noodle Gallery: an open-source, self-hosted alternative to Google Photos and Immich

Noodle Gallery is an open-source, self-hosted photo and video platform its authors present as a Google Photos alternative — more precisely, a fork of Immich with many additions. The project was posted to Hacker News on 21 September, and its code is published under the AGPL-3.0 license.

A fork that stays close to Immich

Gallery calls itself a friendly fork of Immich: it rebases onto every upstream release, so fixes and performance work from the Immich team land here too, and it is currently based on Immich v3.2.2. Moving over from Immich takes three lines of configuration — two image names in docker-compose.yml and IMMICH_VERSION=v5 in the .env file — while the existing database and media library stay compatible. A companion script handles the reverse: it drops Gallery-specific tables and columns and leaves a plain upstream Immich database, and the photos and videos themselves never move.

What the fork adds

Shared Spaces are the headline feature: multi-owner collaborative timelines with Owner, Editor and Viewer roles, a per-space activity log and cross-contributor face recognition, where photos are linked by reference at no extra storage cost. Global People deduplicates people across personal libraries and shared spaces without crossing permission boundaries. On the infrastructure side the fork adds S3-compatible storage (AWS S3, MinIO, Cloudflare R2, Backblaze B2, Wasabi) with a migration tool that resumes after an interruption and can roll back, plus Connected Libraries for linking external folders into shared spaces.

For media handling there is an in-browser import wizard for Google Takeout archives that preserves dates, GPS coordinates, descriptions and album structure; automatic classification of clutter such as screenshots, memes and receipts; memories built from location clusters and birthday photos; pet detection with the RF-DETR model; duplicate video detection that compares CLIP embeddings across sampled frames; non-destructive editing; and lossless video trimming through FFmpeg stream copy. Native iOS and Android apps add background backup, on-device CLIP search and the map view.

How to run it

The installation is a single Docker Compose file, with images published to the GitHub Container Registry (gallery-server and gallery-ml, plus a CUDA build). The project says the whole setup takes about five minutes, and a public demo instance is available. There is no subscription and no storage cap — the cost is the hardware. Documentation lives at docs.opennoodle.de, and the maintainers state openly that the codebase is developed with heavy assistance from AI coding tools, with unit and e2e suites on every pull request.