Self-Play Fine-tuning of Diffusion Models for Text-to-image Generation | Read Paper on Bytez