You can use 512 or 768 image sizes but it seems to work best with 768, that's what all the images i trained were limited to.
Works best with the lanczos upscaler.


A model I made by using all the artwork I ever drew, all the 3d models I ever made, all the images I've generated through stable diffusion with permissions that let me use them how I wanted.


Basically gathered up all the best stuff I had ever put together, built them up in loras then force merged them into the 1.5 model with SuperMerger, however, with my newer computer I've been training them in dreambooth!

I'll be looking into what people use it for and how it's used to keep improving it in the future using the same method.