PH

Photonix

Self-hosted photo gallery with face and location detection

Photo Management ★ 2k stars Medium setup AGPL-3.0

Photonix is a self-hosted photo manager that organizes and tags your collection automatically. It uses AI to recognize objects, faces, and locations and offers powerful filtered search.

Key features

  • AI object and face recognition
  • Location-based browsing
  • Powerful filtered search
  • Multi-user support

Pros & cons

Strengths

  • Strong AI tagging
  • Good search experience

Trade-offs

  • Resource intensive
  • Development can be slow

Photonix replaces

Last reviewed Sep 13, 2026 · 730 words

Photonix is a photo manager I would have recommended in 2021 and cannot recommend today. The idea was right: point it at a folder, let it classify objects, detect faces, read GPS into a map, and filter by colour and style through a fast search bar. The parts that shipped still work. But the repository, at about 1,958 stars, has had very little movement for years, the mobile app never appeared, and the photos category moved on without it. If you are choosing a Google Photos replacement in 2026, start with Immich; read on if you are curious, already running Photonix, or want to know exactly what you would be giving up.

What still works, and works nicely

The search experience was the best part and remains good. Tags come from several small models: an object classifier, a face detector with clustering so you can name people, a colour analyser, a style classifier (macro, aerial, bokeh and so on) and a location tagger fed by EXIF GPS. Everything becomes a filter chip, so "beach, blue, 2019" is three clicks. Photos are read in place from a folder you mount, which suits people whose library is already organised on a NAS and who never wanted a tool that copies files into its own structure. Multiple users can each have their own library. Nothing here is unique any more, but the interface around it is calm and quick.

Why it lost

Three reasons, in order. First, no mobile app and no backup path from a phone, which is the single feature that makes people leave Google Photos and the one Immich built first. Second, the models are frozen at the versions bundled years ago, so object and face recognition is noticeably worse than the CLIP-based search in Immich or PhotoPrism. Third, the maintainers went quiet, so bugs sit and Python dependencies age in place. The Immich vs PhotoPrism comparison is the choice that actually matters now.

If you run it anyway, the 2 GB is real

The stack is Django, a React front end, PostgreSQL and Redis, plus worker processes that run the models. The repository ships an example Compose file and you should copy that one rather than mine, but the lines that matter look like this:

services:
  postgres:
    image: postgres:alpine
    environment:
      - POSTGRES_DB=photonix
      - POSTGRES_USER=postgres
      - POSTGRES_PASSWORD=change-me
    volumes:
      - ./data/postgres:/var/lib/postgresql/data
  redis:
    image: redis:alpine
  photonix:
    image: photonixapp/photonix:latest
    ports:
      - "8888:80"
    environment:
      - POSTGRES_HOST=postgres
      - POSTGRES_DB=photonix
      - POSTGRES_USER=postgres
      - POSTGRES_PASSWORD=change-me
      - REDIS_HOST=redis
      - ALLOWED_HOSTS=*
    volumes:
      - ./data/photos:/data/photos
      - ./data/cache:/data/cache
    depends_on:
      - postgres
      - redis

Pin the Postgres major once you have data, because a major upgrade needs a dump and restore. Bring it up, create the first account with Django's createsuperuser through docker compose exec photonix (the README gives the exact path), then log in on port 8888 and add a library pointing at /data/photos. Budget the full 2 GB of RAM: the models load into memory per worker, and the first pass over a 50,000-photo library pegs a CPU core for a day or more. There is no GPU path. Drop files into the mounted folder and a watcher picks them up; that, plus the web UI, is the entire import story. Back up the Postgres volume with pg_dump and leave the cache out; thumbnails and model output regenerate, the database with your face names and albums does not.

Getting out is trivial, which helps

Because Photonix reads in place, leaving costs nothing: your files never moved. Tags, face names and albums live in Postgres and do not export in a format anything else imports, so treat them as disposable. Point Immich at the same folder as an external library and it will re-index with better models in a few hours, and PhotoPrism can do the same.

What I'd do

New install: skip Photonix, run Immich, and follow the Google Photos exit guide for the migration order. Already on Photonix and happy: keep it off the public internet, pin the image tag so an accidental pull cannot break it, and plan the move for whenever you next touch the box. The design was ahead of its time; the maintenance was not, and in this category that is the whole story.

Compare Photonix

21 head-to-head comparisons.

Similar photo management apps