In-Context Forcing: Uncovering Context Effects in Autoregressive Video Diffusion

2026-08-05

AI Summary

Existing few-step autoregressive video diffusion models rely on fully denoised clean frames as context, which often leads the model to take shortcuts and compromises temporal semantics and dynamic consistency. This work proposes a progressive autoregressive paradigm featuring an “In-Context Forcing” mechanism that leverages context frames with progressively decreasing noise levels to provide adaptive guidance. This approach decouples the strong dependency on preceding clean frames and enables parallel cross-frame denoising. The method significantly improves inference efficiency while maintaining high visual fidelity, achieving superior performance over existing approaches on VBench and enabling a synergistic optimization of generation quality and temporal coherence.

Abstract

Current few-step autoregressive video diffusion models depend on previous fully denoised clean frames as context for all denoising steps of the current frame. However, these clean frames leak excessive local details, which causes the model to take shortcuts, resulting in compromised temporal semantics and dynamics. Inspired by the perspective of diffusion as masking, we explore the impact of noisy contexts on few-step autoregressive generation. Yet, simply applying contexts with the same noise levels provides insufficient guidance, leading to poor temporal consistency. To resolve this dilemma, we introduce In-Context Forcing, a progressive autoregressive paradigm that utilizes contexts with decreasing noise levels. By applying less masking to distant frames and more masking to adjacent ones, this approach provides adaptive guidance, effectively ensuring both robust temporal consistency and high inter-frame dynamics. Furthermore, by decoupling the strict dependence on previous clean frames, our paradigm enables cross-frame parallel denoising, achieving substantial inference acceleration without sacrificing performance. Extensive experiments on VBench demonstrate that our method significantly outperforms state-of-the-art approaches in both visual fidelity and inference speed.

Problem

• autoregressive video diffusion
• context effects
• temporal consistency
• diffusion masking
• few-step generation

Innovation

• In-Context Forcing
• autoregressive video diffusion
• noisy context
• temporal consistency
• parallel denoising