RealCompo: Balancing Realism and Compositionality Improves Text-to-Image Diffusion Models

Diffusion models have achieved remarkable advancements in text-to-image generation. However, existing models still have many difficulties when faced with multiple-object compositional generation.