Temporal state balance improves video generation consistency and quality
Temporal State Transport in Video Generation: Diagnosing and Correcting Spectral Imbalance
Computer Vision and Pattern RecognitionMachine Learning
Summary
Generating videos that look good frame by frame is not enough; the video must also look natural over time, keeping things like characters, movement, and details consistent. The researchers studied how well video models maintain this consistency by analyzing how information flows across frames, identifying two common problems: either the video looks broken up or too mixed together. They developed a way to measure these issues and a fix that adjusts the video generation process without needing extra training. This method helps existing video models create smoother and more coherent videos without changing their original design.
video generationtemporal consistencycross-frame attentionspectral tensionstate transportattention diffusenessspectral diversitytemporal hotspotstraining-free correctionvisual coherence
Authors
Luyao Tang, Bingjun Luo, Dong Yi, Jialin Guo, Haoning Xi, Cheng Chen, Yizhou Yu, Chaoqi Chen
Abstract
Reliable video generation requires more than high-quality frames to form a coherent story: a model must maintain a persistent state, transporting visual attributes such as identity, scene layout, motion, and fine details across time. Existing training-free methods mainly strengthen cross-frame attention or analyze local attention entropy, but these views do not reveal whether temporal interactions stay in a healthy transport regime. In this work, we study video generation through the perspective of Temporal State Transport. We introduce Spectral Tension, a signed diagnostic that compares local attention diffuseness with global spectral diversity, and use it to identify two opposite temporal failures: fragmented transport and over-mixing hotspots. Based on this diagnosis, we propose Spectral Transport Homeostasis, a training-free regulator that softly corrects pathological temporal states while largely preserving balanced ones. Experiments on pretrained video generation models show that the original model often occupies imbalanced temporal regimes, whereas our method selectively applies larger corrections to the worst temporal hotspots and improves temporal consistency and visual quality without finetuning. Code: https://github.com/lytang63/temporal-state-transport