Ah well, I am definitely not at that level yet! ; ) I do find inpainting to be very useful, but the results I get usually mismatch slightly in texture and require a bit of blending or color temperature adjustment, etc., to look natural. Like with
this image, I did use AI to give her a better face than the original face, but there were still tiny flaws in the teeth and nose that I found easiest to just fix directly in photoshop. And the blue glow between her legs was also added by me, as was one of her fingers...I find that for tiny details like that inpainting usually produces unnatural or out-of-place results. But I am still experimenting with all of this stuff!
While it's true that inpainting usually introduces inconsistencies like a tone shift, it's usually small enough to prevent it from being useful. If you're experiencing much trouble doing inpainting, it might be the case that there are some issues in the way you use the feature.
Inpainting can look deceptively simple, but in fact, there are many variables involved in the process that can either make or break the result. Many inexperienced users just mark the area to be changed and hit the generate button, which may not work well in many situations.
For example, if you try to inpaint the girl's face in your example image that way, it probably won't work very well. The reason is that the area is too small, and the girl's pose and expression aren't very common.
The infamous problem with SD failing to depict human hands also stems from the same limitation. In order for SD to generate such details (e.g. fingers) correctly, it needs a "good context" and a large enough area. So, when you just select a very small area, it can't understand how its content is supposed to look in the context of a bigger picture. Most inpainting tools automatically include expanded areas around the selection, but it's often not enough, especially when it's too small.
So, ideally, you'd want to include strategically chosen spots in the context (e.g. those parts that can help AI understand the context, like the pose of the girl in your example), then upscale the area to give AI enough space to work with, and finally resize it back to its original size to paste it over the image. Many AI tools provide ways to do such things without too much hassle, but you have to understand the process to get the most out of it.
To give you a more concrete idea, let's assume you want to inpaint a character's face while preserving the overall tone and lighting condition. What you can do is select a few areas
outside the face, and
deselect a few areas inside it. If you select a few tiny spots outside the face, for example, you are effectively including the bounding area into the context while still restricting the area to change to the face. AI will then use more information to infer the correct perspective/tone/lighting for the face area. Deselecting spots on the face, on the other hand, will tell AI to preserve them as they are, so it will try to find the best matching perspective/tone/lighting for the surrounding area to keep the overall image more consistent.
By the way, you can also draw on control net images, such as one for the lineart or scribble model. For someone like you who has excellent skills in Photoshop, it can be far easier and better to hand draw on control net images to add details or fix flaws. Drawing over generated images can have its uses, but it's best followed by a low-denoise pass for the best outcome.
Sorry for the wall of texts, providing you with unsolicited help. However, I couldn't resist since I've struggled with similar issues for a long time until I gradually obtained a better understanding of the tools.