Loading...
searchstack.dev
Sign In

Image search

Find the records in a list or group that look most like a picture you supply.

  • The picture becomes a numeric fingerprint of what it shows — the same kind text search makes from words — compared against your indexed records’ fingerprints. A photo of a red trainer finds other red trainers even when nothing says “red” or “trainer”.
  • Runs alongside text search; the same list serves both.
  • Powers “search by photo”, “shop the look” and “more like this”.
  • No console screen — see the live image demo.
1

A library with no tags and no captions

Each picture is turned into a numeric fingerprint of what it shows as it is indexed. Nothing here is labelled: there is no alt text, caption or keyword doing the work.

A photo library indexed for image search
2

Give it a picture, get the ones that look like it

Clicking a photo sends that photo as the query, and the closest matches come back ranked, with the similarity score on each. That is the /search/images call below with the image supplied by URL — and the score is what minimum_image_score filters on. Try it on the image demo.

Visually similar photos returned with a similarity score on each
Two ways to supply the image
  • By URL (image_path) — a full, publicly reachable image address. For images already online: a product photo, a picture in your CMS.
  • By base64 (image_base64) — the image file itself, encoded as base64 text. For camera uploads and files you don’t want at a public address.
The request

Every image path lives under /search/images:

GET https://api.searchstack.dev/search/images/{account}/{list}/{version}?image_path={image url}
  • /search/images/group/… searches every list in a group at once.
  • /search/images/base64/… takes the encoded file in a POST body, so the picture never has to be hosted.
  • The body takes the same options as a text search: filter over facet fields, size/skip for paging, and minimum_image_score to drop weak visual matches. The right floor depends on your model and images — start without one, then adjust.
  • The response matches a text search: a ranked results list, closest first, each scored under @image_score with your own fields under fields.

Every path, parameter and response shape is in the API reference under Search.

When to use it
  • “Shop the look” and “find similar”: upload a photo of a chair, find the chairs in your catalogue that look like it.
  • Spotting duplicates before adding a picture to a catalogue.
  • Media libraries whose files have little or no text to search on.
What it needs
  • An embedding model that fingerprints images as well as text (a multimodal model).
  • Records carrying image URLs — typically a resource field pointing at a media store.
  • Index those records once; the same list then answers both text and image queries.
Top