>>3018570If you've got that kind of hardware you might want to see what you can do with Flux based models and/or video gens.
SDXL was original trained at 1024x1024, so that's what it expects whatever concepts in the prompt to fill. You can go higher like 2048x2048, but you might get duplicated body parts and such because the AI thinks a person should fit in 1024x1024.
A better way to get higher res images is with an AI upscale, like the hi-res fix built into A1111/Forge/Reforge. Then you can upscale your base gen to a higher resolution, like going from 1024x1024 to 2048x2048. You could even do multiple upscales since with your GPUs they shouldn't take much time.
You can also look into ADetailer, an automated inpainter:
https://huggingface.co/Bingsu/adetailer . Inpainting is a great way to get better details like for eyes, and ADetailer will try to automatically detect different things you can pick like faces, eyes and hands and inpaint those things for you.
If you have that kind of tech you can probably do really impressive things with it so it'd be neat to hear if you get things working how well it goes for you.
