Skip to main content

AI descriptions

A local model writes a short description of each photo, which makes it searchable, and lets you ask questions about your library. It runs through Ollama on your server, so your photos never leave it.

Turning it on​

  1. Add the profile to .env, together with ml if you use it:

    COMPOSE_PROFILES=ml,ai

  2. Apply it:

    docker compose up -d

  3. Under Admin → Settings → AI descriptions and questions, set the address to http://ollama:11434.

The models take about 3 GB of disk and download by themselves.

Processor use

Without a graphics card, each photo takes a while and Ollama uses every processor core while it works. On a server shared with other services, those will slow down, especially after a large upload.