You know, thinking about it, I doubt this is a coincidence.

The finger-counting is familiar to me as a technique for lucid dreaming. If you look at your hands in a dream, your brain will kinda fuck it up, so if you train yourself to pay attention to that you realize you are dreaming and become lucid.

My guess is that the origin of fae is something like sleep paralysis deamons or hallucinations, and people realized they could detect those from the same flaws of our own imagination.

Now for AI, it isn't really drawing. What we are using in image-AI is still much more like projecting up a mental image, dreaming. We can't get it right all at once either, even our human brain is not good enough at it, it is reasonable image-AI makes the same kind of mistakes.

The next step would logically be to emulate the drawing process. You need to imagine up an image, then observe it at large, check for inconsistencies using reasoning and visual intuition.
Hone in on any problems, stuff that doesn't look right or doesn't make sense. Lines not straight.
Then start reimagining those sections, applying learned techniques and strategies, painter stuff (I am not an artist).
Loosely I imagine the ai operating a digital drawing program with a lot of extra unusual tools like paste imagination or telepathic select, or morph from mind.

The main thing differentiating dreaming from painting is that for painting you can "write stuff down" and don't have to keep it all in your head all the time. This allows you to iterate and focus in without loosing all the detail everywhere else.