I'm looking for a decent ai model that can do image detection, classification, whatever so I can organize the GBs of memes I've collected. I want to do this all locally if possible.

You might be able to find some models on ollama

What @BordedDev says. Look into llama-vision or moondream2 models. It's very doable. The models aren't that big. You could have a working setup within a hour to test if it satisfies your needs.

Edit: check this, that's all: https://ollama.com/blog/...

@retoor I found moondream and it technically did what I'm asking, but I assume it wasn't trained on memes because it was not very good at all at detecting them. I found llava but I'm afraid that's only trained on acceptable use policy friendly stuff too.

I have a 4070ti 12gb if that helps with answering the question.

@cuddlyogre I understand what you mean. They're destroying everything with their censoring. Mine refuses to execute some commands on my system like rm * or smth. If i want that, do that ffs. That stupid thing just tells me what's allowed on my own system.

I see that dolphin vision 72b exists. Dolphin is uncensored pretty good regarding the other type of models. It sadly isn't pushed to ollama tough. In that case you need to find a ollama replacement. @BordedDev knows ways how to run hugging face models in comfortable way.

@retoor yeah, but I think it's easiest to just modify the example snippet they have (I normally use a llama.cpp wrapper that's super nice, but it doesn't support vision at the moment)

@retoor @BordedDev That looks very promising. Thank you very much