Predictive Structure Improves Video Diffusion Dynamics

Learning How the World Evolves: Extrapolative Video World Models via Latent Dynamics Reasoning

The world evolves following its dynamics, i.e., its laws of motion. However, leading video diffusion models largely fit the pixels without modeling how the pixels transit over time. Thus, they render visually plausible frames but may not accurately obey the laws. To capture the dynamics purely from pixels, we introduce Latent Dynamics Reasoning (LDR). LDR casts the latent transition as an explicit kinematic integration, where the lower-order dynamics are integrated numerically and the model regresses only the third- and higher-order residual that drives the rollout. For this integration to extrapolate better, LDR runs it on a structured latent rather than dense convolutional features. Following PhyWorld, we validate LDR on a controlled white-box physics benchmark spanning five tasks (uniform motion, parabola, collision, bouncing, looming), focusing on out-of-distribution scenarios that reveal whether a model has truly learned the underlying dynamics. LDR extrapolates the learned dynamics far better: the gap between its in- and out-of-distribution error is over 20$\times$ smaller than the video diffusion baseline's, under both single- and joint-task training at 256$^2$ resolution, while using 26$\times$ fewer parameters and running 143$\times$ faster. LDR can even generalize under severe shift: for example, trained only on red balls moving left-to-right, it correctly predicts the motion of a blue square moving right-to-left. To our knowledge, this is the first video world model that extrapolates learned dynamics beyond its training distribution. Project page: https://lat-dyn-reason.github.io/

Long-Horizon Video Generation with Temporally Consistent Diffusion and Scene-Graph Guidance

The synthesis of high-fidelity, temporally coherent long-horizon videos remains a profound challenge in the domain of generative artificial intelligence. Current diffusion-based approaches often suffer from severe temporal degradation, semantic drift, and structural inconsistency when generating sequences beyond a few seconds. To address these limitations, this paper introduces a novel framework that integrates temporally consistent diffusion models with dynamic scene-graph guidance. By leveraging scene graphs as explicit semantic anchors across frames, the proposed architecture structurally constrains the generative process, ensuring that objects, their attributes, and their interrelationships remain stable over extended durations. The methodology involves a dual-stream architecture where a graph neural network processes sequential scene graphs to condition a cascaded video diffusion model. Furthermore, a specialized spatiotemporal cross-attention mechanism is developed to align latent noise representations with the relational data embedded in the scene graphs. Extensive empirical evaluations on standard video generation benchmarks demonstrate that the proposed method significantly outperforms baseline approaches in both quantitative metrics and qualitative human assessments, particularly in maintaining entity persistence and logical action progression over long time horizons. The findings underscore the critical role of explicit structural representations in overcoming the inherent memory limitations of pure attention-based video generation systems

Gen4U: Unifying Video Generation and Understanding via Diffusion

Gen4U (Generation for Understanding), a framework that repurposes these generative representations with a single forward pass, is introduced, achieving strong perception performance while fully preserving the model's ability to generate high-quality video.

Cycle-World: Mitigating Error Accumulation in Long-term Video World Models via Reverse-Prediction Cycle Consistency

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.

Self-Supervised Learning of Structured Dynamics from Videos

The Structured Dynamics Model (SDM) is proposed, which explicitly separates the dominant source of temporal change from residual dynamics through future-feature prediction, rather than representing video change with a single entangled latent or with unstructured, spatially dense transition tokens.

Streaming Diffusion Model for Fast Infrared and Visible Video Fusion