>>94099276For the inpaint try adding the "differential diffusion" node before the inpaint ksampler, it sort of helps better understand area context and meld stuff better so may help, other things to try, I notice when getting context area with "ranged size" it tends to extend outside the available image so you get wonky context, try using "forced size" and adjust context expand size as necessary you shouldn't for the most part need to adjust prompt if you are only fixing stuff in a image. Yeah inpaint is gonna be the biggest difference you will feel from NAI since they have a inpaint model, I use Flux's inpaint model, I can see how powerful one of those is (was hoping noob guys would train one kek)
For the upscale, I have not used that Ultimate SD upscaler that much but I think what that does is split the image into smaller chunks and upscales each portion individually then combines it at the end. So if you get a lot of artifact I would assume you either want to increase size of the chunks so it has more context as it upscales or lower the denoise.
