Yep, especially using something called "ControlNet", you can give it a base image, then tell the AI what to generate. It'll use the outline (or other properties) of the image to "guide" the image it generates, resulting in images like this one.

Regarding the comment about fingers, take a closer look at the architecture and especially the text/images on the signs. AI still has a ways to go, though it's getting better by the day.