[Feature] Search by selected image crop / region #28463
johnny-salz
started this conversation in
Feature Request
Replies: 1 comment 4 replies
|
cool idea. just wondering: did the "view similar photos" feature not satisfy your result expectations? I suppose specific details in a "bird-eye" shot are hard to find matches on, right? |
4 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment



Uh oh!
There was an error while loading. Please reload this page.
I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.
The feature
I would like Immich to support searching by a selected region of an image, conceptually similar to how text query works but with image part.
Use case: I open a photo, draw a rectangle around some object or detail, and search for visually similar assets based only on that crop. This is useful when the full image has too much unrelated context, but one object in it is the thing I want to find elsewhere.
I made a local PoC and it seems fairly easy to wire into the existing smart search flow:
In the photo viewer, add a “Search crop” action.
Let the user draw a rectangular crop over the displayed image.
Send normalized crop coordinates plus assetId to the server.
Server reads an existing preview/thumbnail for the asset, extracts the crop with sharp, and sends that temporary image to the existing machine learning clip.visual endpoint.
Use the returned embedding with the normal smart search vector query.
Return the usual search results.
I have a working local PoC if maintainers want to see the patch or discuss the API shape.
search:



crop results:
vs full asset results:
Platform
All reactions