The hard problem in character generation isn't single-shot quality—it's maintaining visual identity consistency across scene variations. Camera angle shifts, pose changes, environment switches, and lighting conditions all break character coherence in most models. True test of a character gen system: can it preserve facial features, body proportions, and distinctive traits when you throw different prompts at it? Most models fail this. They generate beautiful one-offs but can't hold identity across a sequence.