With all the prompt crafters around (it seems a new profession or skillset is rising), I wondered if the obvious problems of AI art could be solved with a little photo editing instead of an endless loop of inputting and outputting images and text prompts.
The most obvious error in AI art is anatomy. Hands and faces, mostly. So I attempted a quick fix and ended with the result above. The prompt used was "Galadriel Cara Delavigne Michael Whelan" *(this[1] is the generator that I used).
Using a famous person as a model means that if the face is too mangled in the result, I can pull the person's images as a fix. Don't know the legal ramifications of this, but I wouldn't risk any actual paid work on it.
Michael Whelan was simply because I love Michael Whelan's work. Obvious legal shenanigans there too. And Galadriel because I wanted some elf/queen/sorceress thing.
Painterly images are easier to edit without looking too much out of place. In my interpretation, it's all right, but I'm sure a lot of people can find a bunch of mistakes.
The only (potentially commercial) use I can find for this is concept art, since it by itself is not the final product. My use for it is essentially goofing off.
Here I attempted to fix Dark Wizard Ozzy's hands. Still looks odd:
