I don't think it's solely an data issue. Flux models for example are quite stylized, very notable with photorealism. But I think it was an deliberate choice to to have outputs that are absent of likeness and distinct style. I think it's an side effect that it washes away fine details and creates outputs feel artificial. The problem is that closed models can't be fixed easily, while models like flux or even older architectures can add back details and style with fine tuning and LoRas.