Lanterns of course.
Would take an image generation model at least 3 steps which it doesn’t have right now.
A review step to see if the output matches the prompt.
A identification step to detect elements that don't match
A redo step to mix that area in the background image (remove) or regenerates an improvement.
Right now you cant iterate on images. Every minor tweak is a completely new image. At least not with dalle because you cant control the seed.
More modern but still not quite reality for many.
If you categorize it as anything then Neurodivergent is the most appropriate label.
Thats not a medical label anymore though but in a medical context you should be more specific anyway.