F-Droid

Like others suggested, immich might be able to solve the use case - it just means you have to sync your photos to immich for the server to index the photos. Then in the app you can search for any recognized objects / locations / etc so from your phone it works about the same as Google photos search. Like I can search beach or cat or whatever from my phone and find all of those photos, and it's private since it's self hosted.

I run immich on my nas so it does both photo backup and provides a nice high performance searchable app / web UI to browse and organize photos into albums. It also does private face recognition and tags people, which are searchable too (the most recent app update actually just brought advanced search to the mobile app) so you can select john doe under the people filters and search fish and up comes only pictures of john showing off fish he caught.

My understanding is that apple/google do that server side because its difficult to do on the phone.

FOSS wise there are a few tools out there but they are self hosted (ex. Immich). Don't know if any app-only version exists yet

ente is an e2ee photo storage that has on-device ML. The AI is "ok." It can generally find what I'm looking for with some false positives thrown in. Not as good as Google's super computers, but it runs on device and is fully open-source, so I'm happy to use it.

I wish these solutions would save their tags to the images themselves. Every gallery with automated or manual tagging seems to have its own way of storing the tags, making them relevant only for that app.

You could try https://f-droid.org/packages/com.slavabarkov.tidy/

the closest I've (just) found is Tidy: https://github.com/slavabarkov/tidy

There is a self hosted option u can run on a server that uses ai embeds to tag things can do video and images I forget what its called tho and its not an app unfortunatly but with some tactful bodging should be doable.

EDIT: turns out there is an app.