>>49203278It's not about being averse to controlnet and inpainting. Half of the reasons why shit looks bad on first gen, even with hires fix or multidiffusion, is down to model issues.
A good example of this is how bad {{{bigger}}} gens looked before the advent of the {{{bigger}}} models and model merges. (I'm looking at you NAI fats, see: skindentation) The only way to get good looking {{{bigger}}} without a lot of luck was with img2img. Today, a lot of those modern {{{bigger}}} models give good output with SAG and Dynamic CFG enabled, so I find myself having 3-4 good images out of a six image batch that look good. I don't bother with hands because that is just so austismo, when my model already does hands that are good enough. Like I could care less if there are 1 more or 1 less finger, or the hands don't look right as long as it isn't distracting from the rest of the image, which is about my whole philosophy when it comes to AI genned anatomy in general.