Trained on 20 images of the same face. Works best with realistic style prompts.