> Because it’s trained on “real” people, will it be easier to generate ugly people? In the literature, testing concepts in image generation is asking human gra

> Because it’s trained on “real” people, will it be easier to generate ugly people? In the literature, testing concepts in image generation is asking human graders "which image do you prefer more for this caption?," so the answer is probably no. You could speculate on all the approaches that would help this system learn the concept "ugly," and they would probably work, but it would be hard to measure.