Researchers at New York University have developed a new architecture for diffusion models that improves the semantic representation of the images they generate. “Diffusion Transformer with ...