Hey, uploader here. I made a mistake. I mean the Maxwell architecture, not the Pascal architecture. My bad! It has something to do with Cuda 5.x being converted differently, leading to a different starting sampling picture. Anything _starting_ from the Pascel architecture mostly results in the same thing, and I can't wait to try it out when I get my hands on it!
Absolutely no way, this looks amazing! You took my prompt and just made it absolutely beautiful, I was really struggling with it, at some point it just started degrading and I couldn't find the root of it. Love it, great model, good tweaks, and most of all, nice lack of double ears.
Absolutely no way, this looks amazing! You took my prompt and just made it absolutely beautiful, I was really struggling with it, at some point it just started degrading and I couldn't find the root of it. Love it, great model, good tweaks, and most of all, nice lack of double ears.
Thanks! I saw your commentary about an outdated berrymix and wondered what it would look like with a different model. I really think you should consider upgrading models- a lot of the newer ones are really quite better. Also with negative embeddings you don't really have to go with the giant negative prompt anymore. You can generally just throw one of them in there. The one I used was https://huggingface.co/datasets/gsdf/EasyNegative, but there are a lot.
