causal-diffusion-transformers

This paper introduces Causal-rCM, a unified teacher-forcing and self-forcing framework for autoregressive diffusion distillation in streaming video generation and interactive world models, achieving state-of-the-art performance with fast convergence.