Words and meaning

What a typed search matches, how meaning joins in, and the AI switch that decides.

A typed search looks for your words in what each photo carries, and, on a plan with vector search, for what your words mean. Both happen in one request to your own Opensolr Index.

01 · Words

Your words are matched against these, from the strongest to the weakest:

  1. The text printed in the photo: a receipt, a label, a screenshot. See printed text.
  2. The names of the people on it. See people and faces.
  3. Your own tags.
  4. What the photo shows, in the words Opensolr read it into, or your own wording.
  5. The camera and the place.

Accents do not matter (montreal finds Montréal), and words are matched as written, without cutting endings off. How many of your words must be in a photo is yours to set: Lexical matching in search tuning.

02 · Meaning

On a plan with vector search, the app sends your words and the search it built to Opensolr's photos_select. There your words become a query vector, and the search runs on your index with words and meaning blended in Opensolr's own hybrid ranking, the same one site search uses. Only the photos found come back; the vector never reaches the phone.

So puppy can find a photo read as dog.

How much the words weigh against the meaning is the Semantic ↔ Lexical Balance in search tuning. It starts at 0.2, mostly meaning.

A search by meaning uses a tenth of an AI request. The same words searched again are answered from Opensolr's cache and cost nothing. See plan limits.

03 · The AI switch

Once something is typed, an AI switch appears next to the count above the grid.

  • On: meaning and words together.
  • Off: words only. Use it for an exact code, a receipt number or a product reference, where meaning would pull in photos that only look alike.
  • Greyed out: your plan has no vector search, or this month's AI requests are used up. Search then matches words only.
04 · When the search runs on words only

The same search runs on the words alone, straight on your index, when:

  • the plan has no vector search, or this month's AI requests are used up;
  • the AI switch is off;
  • what is left to search by meaning is a single character;
  • Opensolr cannot make the vector at that moment. Nothing on screen says so; the next search tries again.

Typing is trimmed to 300 characters. A search runs when you press Enter, pick a suggestion, clear the box or change a filter; typing alone never changes the grid.

Finding photos

Opensolr Photos is open source and MIT licensed. Questions about your Opensolr account, index or plan go to opensolr.com/contact; questions about the app itself belong on GitHub.

Opensolr Photos Documentation