@brianjesse yes we have one feeding memes into the Fediverse. I'll let you talk with my co-conspirator @rtp
But we have a workflow that includes a rudimentary AI that analyzes the images and tries to describe it. A work in progress that has evolved a lot.
@brianjesse I tried LLaVA a few month back and was rather positively surprised. It's not too demanding on resources but still too much to host it on a server. I think I've seen some ideas to build llama.cpp to WASM, so maybe that would be an option.
It can anyways become a great tool for people with impaired vision.