Ziheng Ouyang

Individual token selection and metadata caching unlock sparse attention for diffusion transformers without sacrificing generation quality, achieving up to $2.32\times$ speedups.

Achieving high-fidelity stylized 3D asset generation in under 10 seconds could revolutionize content creation in gaming and animation.

Diffusion models can now capture nuanced semantic and material details in image stylization, moving beyond simple color-driven transformations, thanks to a Mixture of Experts architecture.