I agree. One thing I've noticed with most image generation, including faces, is that it makes the same mistakes that a novice artist makes. It draws eyes where eyes should be, mouths where mouths should be, but doesn't actually shape them correctly for the 3D head and viewer perspective. It can perfect a specific viewpoint with lots of training (like a novice artist who perfects drawing realistic faces and not much else). But my interpretation of much of the artifacts are that they are just missing structural cues.