It works better on anime models, although I think you can achieve results with realistic models by writing more correct prompts