The Seriality Gap in Video Diffusion Models
2026-07-14 • Machine Learning
Machine LearningComputer Vision and Pattern Recognition
AI summaryⓘ
The authors studied how video models predict what happens when one ball hits another in a sequence of hits. They found that standard video diffusion models perform worse as the number of hits in a row increases, and this isn't just because the videos are longer. Their experiments show that the problem comes from the model struggling with tasks that need step-by-step reasoning, not just longer videos. They call this the 'seriality gap' and prove that simply adding more denoising steps won't fix this issue for tasks needing serial thinking.
video predictiondiffusion modelshard-sphere dynamicsserial computationautoregressive generationdenoising stepscausal chainblockwise generationvideo simulationdeterministic prediction
Authors
Jorge Diaz Chao, Konpat Preechakul, Yuxi Liu, Yutong Bai
Abstract
When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.