Mu Yang

NeuWorld achieves impressive long-horizon consistency in interactive video generation without relying on pretrained models or auxiliary 3D reconstruction methods.

Forget slow, view-dependent 2D diffusion: this work generates consistent 3D scenes directly from a learned 3D latent space, unlocking faster and more controllable scene generation.

Double the emotion conversion accuracy in voice conversion models with a simple prefix that jointly controls sequence modulation and acoustic realization.