>>11307635
Yeah I was finding even 1 point could mean it would completely reinterpret the location and position of an arm, which was kind of bizarre.

>I found that there's no need to be overly detailed
I'll figure this out better over time but I have a perfectionist tendency to at least make sure nothing is there I don't want being passed through the gen.
Cuz when refining details, even a few pixels of white on a lip can make it overemphasize that and eventually grow into parts of teeth and other bullshit.

Maybe it's just precision, that I'm trying to aim for, even with the barely-better-than-stick-figures starting image. The thought there is that I've found it nearly impossible to tell the model something is there that isn't.
i.e. the image has a blue jacket and you want it to be red
No amount of inpainting or prompting has gotten me far enough in that kind of scenario.
But yeah my wrist did not like making that base image so some kind of tablet/pen option sounds really nice on my end longer term.

Also, how do you approach resolution?
I'm curious if you have a (or multiple) go-to res for the base painting and maybe have a goal to upscale to by the end.
I think the biggest issue I'm having right at this moment is the base image was pretty low res and fitting in detail cleanup through inpainting isn't as successful as I'd like because of it.

I decided to toss it in img2img with as much of the original prompt as possible and upscaled it by 1.5 from 848x1120 to 1272x1680 and re-touched it up all over again.
(initial res was definitely too small for this)