Search by Image Now Reads Labels: Photograph a Shelf Tag, Get the Exact Product

· Search · Improvement · All updates

Two days ago you could search with a photo of a thing. Now you can search with a photo of a label. Point your phone at a shelf tag in a store, and the exact product it belongs to comes back first — because Opensolr read the tag, instead of just recognising that it was one.

01 The photo that started it

A shelf tag on an interior door in a DIY store: a spec table, a model code, an article number, a barcode. Before today, this picture became Label, Document, Paper and found nothing useful. Now it becomes USA LEMN G1-88 E · 6048297, and the first result is that door.

02 Recognise it, or read it

Every photo still goes through the visual step first: Opensolr picks the words that describe what is in the picture, from a vocabulary that grew from 21,000 to 51,000 concepts today — species, breeds, dishes, product categories. When those words say the picture is a piece of text, it takes the second path and reads it.

Your photo a thing, or a label A thing: recognised husky, sunflower, faucet A label: read name + article number Your results your search runs
One photo, two ways to turn it into words. Your index does not change for either.

03 What changes on your search page

  • The page tells you what it did. Under the search box: Showing results for your image: sled dog, mushing… for a picture of a thing, Read from your image: USA LEMN G1-88 E · 6048297 for a label. It shows exactly what was read, misreads included, so you always know why the results are what they are. Long lists fold behind a show all.
  • Labels search exact, automatically. An article number is something to match, not something to understand, so a label search runs with the AI toggle off — and the toggle goes back to your own setting on the next thing you type.
  • Bigger photos. The upload limit is now 20 MB, enough for a full-size shot from a modern camera. HEIC from iPhones is accepted.
  • Nothing to switch on. Every hosted search page has it already. Facets, filters, AI Hints and everything else keep working on top of it.
Works best on pictures that are mostly text: shelf tags, spec labels, receipts, documents, posters. A product with a small label on it is still recognised as the product — the label is read only when it is the subject of the photo.
View the full changelog