I don’t think that fact has been established. From what I’ve observed with Midjourney and Dall-E, the defects in the output of these models are more artisticall

I don’t think that fact has been established. From what I’ve observed with Midjourney and Dall-E, the defects in the output of these models are more artistically significant than just mangled faces and missing details. The models don’t understand the styles they are aping, and they don’t understand the elements of visual art in a way that allows them use those elements in the service of an expressive goal. I think it’s quite plausible that we plateau to a point where the user of these systems still needs significant technical knowledge (drawing and painting) in order to correctly specify the requirements to the model.