>>1455
Is DeepBooru the same as Deep Danbooru, or the same basic model, so you could switch it in place? Hydrus-dd offers several ways to plug into Deep Danbooru, here:

https://gitgud.io/koto/hydrus-dd

I was talking with someone else today about how I'd like to do more with this sort of tech in future, just making it easier to work with the sidecars you need or the Client API whatever. I'm sure image-to-text is going to benefit from all this text-to-image tech that's now floating around, especially now it is becoming reasonable to train your own models to do stuff. Hydrus as a big metadata-rich archive seems like it plugs into that tech pretty well!