Feature Reuqest: Natural language image search using CLIP Embeddings

[clawoflight]

Describe the feature you'd like to request

I would love to see natural language image search. You might know it from google photos or iCloud - searches like "A dog and a girl in snow" or "girl on beach with sunset".

Describe the solution you'd like

The industry standard seems to be CLIP embeddings, as e.g. implemented in immich:
https://docs.immich.app/features/searching/

Describe alternatives you've considered

The existing tag-based search is inherently limited and cannot cover the same use cases. It is also less accurate.

[marcelklehr]

Hello @clawoflight
Thank you for taking the time to give feedback! Indeed we have been considering adding support for CLIP embeddings to nextcloud, however, we believe that the context chat app lends itself more to this endeavor than the recognize app, as context chat already ships with a vectorDb and has a vector search feature.