Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates the limitations of standard bidirectional video diffusion models in predicting the outcomes of sequential events, revealing that performance declines as the length of causal chains increases. Through controlled experiments, the authors demonstrate that this degradation is due to the dependent-event structure rather than simply the length of the video, as evidenced by improved performance in single-ball scenarios. They introduce the concept of the "seriality gap," highlighting that existing denoising methods fail to provide the necessary serial computation for effective reasoning in video prediction tasks, thus identifying a fundamental obstacle for these models in handling complex temporal dynamics.
Video diffusion models struggle with causal reasoning, showing a significant performance drop as the complexity of event sequences increases.
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.