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
Autoregressive video generation has emerged as a powerful paradigm for World Action Models (WAMs). However, existing approaches suffer from slow training convergence and limited converged accuracy, particularly at high frame rates, as the training supervision is confined to the current chunk without explicit signals about future dynamics; they also suffer from slow inference due to iterative video denoising.
The paper introduces Relax Forcing, a training‑free memory mechanism for autoregressive video diffusion that structures temporal context into Sink, Tail, and History frames. By selecting History frames with a relaxation criterion, the method reduces error accumulation and attention overhead while preserving motion dynamics. Experiments on VBench‑Long demonstrate that this structured memory improves long‑video generation quality over existing baselines.
By Zengqun Zhao, Yanzuo Lu, Ziquan Liu, Jifei Song, Jiankang Deng, Ioannis Patras
Autoregressive video diffusion models have emerged as a promising approach for long video generation, achieving strong performance in streaming settings. However, existing methods are restricted to forward temporal generation, whereas practical video creation often requires flexible generation order, e.
arXiv:2610.03123v1 Announce Type: cross
Abstract: Existing streaming vision-language models (VLMs) continuously perceive and reason over visual streams, but their computational pathways remain fixed...
By Ashok Prasad Neupane, Dipan Bartaula, Ankit Belbase, Saugat Adhikari, Samip Ghimire, Saroj Poudel, Binod Bhattarai, Danda Pani Paudel
arXiv:2609.39563v1 Announce Type: new
Abstract: Existing video language models encode sampled RGB frames independently, so a long video must either exhaust the token budget or drop the changes betwee...
By Can Zhang, Xiaotian Han, Junyuan Shang, Yuchen Ding, Zhenyu Zhang, Shuohuan Wang, Dianhai Yu, Ruirui Li
arXiv:2610.33589v1 Announce Type: new
Abstract: While interactive video world models require autoregressive generation, directly adapting pretrained bidirectional video diffusion models into autoregr...
By Zelin Gao, Hao Ouyang, Jiapeng Zhu, Qiuyu Wang, Jingye Chen, Zichen Liu, Xing Zhu, Yujun Shen, Yinghao Xu
MT‑WAM enhances the Fast‑WAM framework by adding complementary supervision for future 2‑D point trajectories and visual features while keeping the original training objectives. A lightweight dual‑stream branch and structured attention mask isolate motion‑specific processing, and motion‑stream tokens provide additional dynamics cues to the action expert. During inference, MT‑WAM skips future‑video prediction, using cached video and motion information to achieve higher success rates on LIBERO, LIBERO‑Plus, RoboTwin 2.0 Clean2Rand, and several real‑world tasks.
By Yiguang Yang, Jiankun Peng, Xiaoming Wang, Yiran Zhang, Zhibo Fang
Direct Forcing is a single‑stage training framework for few‑step autoregressive text‑to‑video generation that eliminates the need for separate causal adaptation and distribution matching stages. It uses reference videos both as paired examples for causal continuation alignment and as a target distribution for rollout distribution matching, with dynamic gradient balancing to adjust their contributions during training. Starting from a bidirectional Wan2.1‑1.3B model, Direct Forcing trains a four‑step generator in only 250 updates, achieving state‑of‑the‑art VBench scores without additional initialization or diffusion‑score supervision.
By Haojun Xu, Jie Huang, Xin Lu, Mingchen Zhong, Zihao Fan, Linjiang Huang, Si Liu
arXiv:2607. 18042v1 Announce Type: cross Abstract: End-to-end vision-language navigation (VLN) with causal vision-language models can map instructions and egocentric observations directly to actions, but standard behavior cloning supervises only the next action and does not explicitly train the policy state to be predictive of future visual outcomes.
By Lingfeng Zhang, Zhanguang Zhang, Liheng Ma, Tongtong Cao, Yingxue Zhang
arXiv:2610.30861v2 Announce Type: replace-cross
Abstract: Few-step autoregressive video generation enables low-latency streaming, yet prevailing training pipelines separate causal adaptation from dis...
By Haojun Xu, Jie Huang, Chuyang Zhao, Xin Lu, Mingchen Zhong, Zihao Fan, Haoyang Huang, Linjiang Huang, Si Liu
ViRDM is a new post‑training method for few‑step causal video generation that eliminates the need for a large teacher model and an online critic. By applying representation distribution matching (RDM) with a precomputed target distribution, a lightweight VAE decoder, and staged vector–Jacobian products, ViRDM overcomes memory, optimization, and temporal dynamics challenges. The approach reduces GPU memory usage and training time, achieving state‑of‑the‑art VBench performance with only 20 generator updates and 16 A100 GPU‑hours.
By Zichong Meng, Chongjian Ge, Chun-Hao P. Huang, Yang Zhou, Huaizu Jiang