General method for fixing some poor area in generated images

[R-N]

Sometimes you generate an almost perfect picture but is poor in some area, such as hands, so you'd want to fix it. Everything else was perfect and you want to have as minimal change to them as possible. What's the general go-to method to do this? By general I meant generally applicable for any model/artstyle.

I've tried simply sending it to img2img and use the same prompt. Of course, it changes the whole thing. Sometimes it does fix the earlier poor area, but it also changed the rest too much that it's no longer the same style, or introduced new poor area. I'm using the same parameters as I did in txt2img.

I've tried inpainting but it's pretty bad. I masked the hand and it gets worse instead, hardly resembles hands. I tried masking the parts I want to keep and let it fill the rest instead, but they dont mix well, at all. Note that I'm not using the SD inpainting model. I'm using NAI.

I tried the loopback script but it's just as previous img2img (not inpainting) and changes the style and form rather drastically. I'm using the default settings.

[cmp-nct]

masked Inpainting with the 1.5 inpainting model can work quite well.
I personally use photoshop to fix mistakes and if necessary any changes can be integrated seamlessly with a low denoised in img2img

[R-N]

Does the 1.5 inpainting model work for anime-ish image? Won't it inpaint with realistic hands instead?