[jwells89]

And this will continue to be an issue until ML models have achieved something resembling sentience, because many of these tells are the result of the model not truly comprehending the subject matter and thus struggling to maintain internal consistency in everything from geometry and kinetics of human bodies to lighting and physics.

Less obviously, ML models also lack the ability to bake in intent. In human made pieces, everything is as it is for a reason; it’s communicating something. In ML generated pieces, things are the way they are because that’s what’s statistically likely for the type of generated image.

[PixyMisa]

Absolutely correct. This is a far more obvious problem in text models, because you end up with internally flawless arguments as to why your next scuba diving vacation should be in Ulan Bator.

With art, it's more subtle, because there there's no single reference point that lets us determine if an artwork is "true". There are the glaring errors that everyone can agree on - notoriously, human hands - but those cases are improving rapidly.