Rex

Thanks for bringing this up, since I was also training 3 different sets one by one in small chunks, and didn't think about the last trained bias. Now really need...

Another thing to consider is that the outputs change with this. I tried it for scoring images, and outputs changed, so I had to revert, because that introduced another variable...

This model was trained using images from 3 sources that were scored by people, so the model uses image path and rating for retraining. For just evaluating your gens, you...

They do in this blog. https://laion.ai/blog/laion-aesthetics/

This is amazing! Thanks so much. Used your base code to create a whole other way to gen images, then score them, then classify them, making incredible art in bulk.

I noticed this as well a couple days ago. I cannot replicate old prompts-seeds. I use my local setup that I have not updated in weeks, and an online version...

Looks like sometimes it makes the image better, sometimes worse.

Pre processing text can help Focus too. Transforming English to simple English inherently cuts the vocabulary from 170K to 2.5K, while not losing context, making it easier for the model...

You are amazing!

FastAPI is easier to work with since you can use Swagger to confirm everything works, and the spec gets created automatically. Now if they just fixed localhost plugins, I would...