For example, the person in the photo in excellent detail on the AI model will be like plasticine.

I wonder if that is deliberate, to avoid lawsuits from celebrities. I'm sure it recognizes that it is a person and then uses different settings so it doesn't create exact matching faces.
Maybe changing the settings will get a better result, for example increasing the sample steps. Or changing from stochastic to multidiffusion (I have no idea what those mean [:))] )