AI acts like an alien more than a copier in that case. If you tell stable diffusion to make you clip-art of people in a conference room, you will probably see humans that don't have noses or fingers, or have 4 arms and some unreadable text that looks like a watermark going across them. The AI parameters have been trained on millions of clip-art images and assume the image should have whatever statistically applies to most images that have the same keywords on them. You can't make it copy an image, even if you tried. You can't even get it to fix the face or hands without additional processing with differently trained models. It sees as an alien would.
