In-Context Forcing: A New Autoregressive Paradigm for Video Diffusion

A new research paper on arXiv (2608.05237) introduces In-Context Forcing, a progressive autoregressive paradigm for few-step video diffusion models. The method addresses the issue of leaking excessive local details from clean context frames, which compromises temporal semantics and dynamics. By using contexts with decreasing noise levels—less masking for distant frames and more for adjacent ones—the approach provides adaptive guidance, ensuring robust temporal consistency and high inter-frame dynamics. The paper is authored by researchers and was announced as a cross-type submission on arXiv.