Enhancing Prompt Understanding of Text-to-Image Diffusion Models with LLMs (bair.berkeley.edu)