>>86486456>or some dumb reason inpainting and img2img share all their settingsWell, inpainting IS img2img and multidiffusion is not much different than selecting a large section of the image and inpainting it.
The inpainting step can be used for big or small changes and really unless it's a very crude edit it's better to do it at the final resolution instead of before upscale. And the final inpainting passes can add a lot of detail especially if you're doing "inpaint only masked" in small sections.
For example, on this I made manual edits in krita, then refined with inpaint at like 0.7 ish denoising, then refined further at lower strengths from 0.5 - 0.6 to bring in final details. At higher denoising it's important to make a big enough selection or add enough padding for it to have good context on what it's generating. Otherwise you'll struggle with stuff like faces and stuff where they shouldn't be. Generally speaking you don't have to change the prompt from the initial one if there's enough surrounding context.