>>67267382I'm still experimenting with my workflow. But right now I gen an image with EF+HLL for the composition/sex/expressions. Inpaint if needed. Upscale with based68 model using the same prompt + whatever style loras I'm feeling. This gets rid of the inherent style that EF+HLL has (like the heavy zankuro influence). And sometimes if I'm going for a specific artist look, I switch to an anime model and do another img2img pass since my style loras have not been retrained for furry models. I did this will all the images I posted yesterday because I was just using a single style lora instead of a mix.
I also think genning with a white background helps it look clean, and I don't have to do more inpainting to fix nonsensical AI backgrounds. Although sometimes I get artifacting in the white area like picrel, but that's way easier to fix.
