Cycle-World: Mitigating Error Accumulation in Long-term Video World Models via Reverse-Prediction Cycle Consistency
2026-07-13 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors address the problem of error buildup in long video generation using autoregressive diffusion models, which causes videos to look worse over time. They introduce Cycle-World, a method that makes the video generation process reversible, meaning the model learns to go forward and backward in time consistently. During training, this reversibility helps the model avoid drifting away from realistic videos, and during generation, the backward model corrects errors on the fly. Their experiments show that this approach improves video quality and stability, especially for long videos up to 60 seconds.
autoregressive diffusion modelserror accumulationlong-horizon video synthesistemporal consistencycycle-consistencyreverse-prediction modelgradient-based correctionlatent representationsvideo manifoldVBench benchmark
Authors
Zihan Su, Teng Hu, Jiangning Zhang, Ruiyan Wang, Ran Yi, Lizhuang Ma, Dacheng Tao
Abstract
Autoregressive diffusion models have enabled high-quality video generation, yet their sequential nature inherently suffers from error accumulation. In long-horizon video synthesis, minor prediction deviations compound over time, inevitably leading to unconstrained generative drift, structural collapse, and severe visual degradation. To address this, we propose Cycle-World, a novel framework designed for stable and temporally consistent long-video generation. Our approach tackles error drift by enforcing strict temporal reversibility across both the training and inference phases. Theoretically, we demonstrate that forward generative drift can be strictly bottlenecked by a cycle-consistency objective. During training, we integrate an efficient reverse-prediction model to implicitly embed causal constraints into the forward generator, compelling it to produce reversible sequences that tightly adhere to the natural video manifold. At inference time, we repurpose this frozen reverse model as a runtime corrector. Through gradient-based cycle guidance, it iteratively refines the generated latent representations, actively suppressing accumulated errors before they are committed to the historical context. Extensive experiments on the VBench benchmark demonstrate that Cycle-World's dual-phase synergy significantly mitigates error drift, achieving state-of-the-art overall generation quality and long-horizon temporal consistency in 60-second synthesis.