arXiv Computer Vision By Bohan Wang, Shuo Chen, Yunzhi Li, Zhaozheng Chen, Junzhe Zhang, Hanwang Zhang

Trajectory Forcing: Exploiting Diffusion Trajectories for Autoregressive Long Video Generation

Read the original on arXiv Computer Vision →

The paper introduces Trajectory Forcing (TraF), an autoregressive diffusion method that uses the denoising trajectories of previously generated video chunks as context for generating longer videos. TraF is trained in two stages: first with ground‑truth videos using pseudo‑trajectories, then with a teacher model that supplies real trajectories from the student’s own rollout. On the 5‑second VBench benchmark, TraF outperforms Self Forcing and Causal Forcing, and it achieves the highest overall quality for 30‑second video generation, extending the training horizon sixfold.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computer Vision.

arXiv AI
Oct 9

Conditional Residual Prediction: Improving Autoregressive Video Diffusion without a Bidirectional Teacher

The paper introduces Conditional Residual Prediction (CRP), a technique that reduces a causal video diffusion model’s dependence on past frames by having the model first predict the current frame without conditioning on history and then adding a residual based on the past. This approach allows training a causal model from an image‑model initialization without requiring a bidirectional teacher or complex distillation, and it nearly closes a 6.14‑point quality gap to a bidirectional model. Using CRP, the authors train Optica, a 2‑billion‑parameter causal video model that generates 5‑second 480p videos and achieves an 82.78 score on VBench with only about 15 million training videos.

By Bowen Zheng, Zhiguang Liu, Jiarong Ou, Rui Chen, Tianyang Hu
arXiv AI
Sep 25

Accelerating Video Diffusion via Training-Free Trajectory Routing

The paper introduces TRACK, a training‑free trajectory routing method that accelerates video diffusion by selectively switching between large and small models during denoising steps. A calibration process generates a disagreement score map, guiding the selection of the appropriate model at each step to maintain quality while reducing computational cost. Experiments on Wan 2.1, Cosmos 3, TurboDiffusion, and FastVideo show speedups ranging from 1.95× to 2.73× with comparable quality and diversity.

By Mustafa Munir, Huy Vu, Shreyas Misra, Rohit Jena, Sajad Norouzi, Ali Taghibakhshi, Anis Ahmad, Anjul Patney, Pavlo Molchanov, Nima Tajbakhsh