pith. sign in

arxiv: 2605.15141 · v3 · pith:G5OT2NBTnew · submitted 2026-05-14 · 💻 cs.CV

Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation

Min Zhao , Hongzhou Zhu , Kaiwen Zheng , Zihan Zhou , Bokai Yan , Xinyuan Li , Xiao Yang , Chongxuan Li
show 1 more author
Jun Zhu
This is my paper

Pith reviewed 2026-06-30 20:57 UTC · model grok-4.3

classification 💻 cs.CV
keywords autoregressive diffusion distillationcausal consistency distillationfew-step video generationframe-wise autoregressionreal-time interactive videodiffusion models
0
0 comments X

The pith

Causal Forcing++ initializes few-step autoregressive diffusion students with causal consistency distillation to support frame-wise 1-2 step video generation.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper targets real-time interactive video generation by pushing autoregressive diffusion models into an aggressive frame-wise regime that uses only one or two sampling steps per frame. Prior initialization strategies for the student model either misalign with the target distribution, fail at few-step sampling, or require prohibitive computation to scale. Causal Forcing++ replaces causal ODE distillation with causal consistency distillation, which supplies supervision via a single online teacher ODE step between adjacent timesteps instead of precomputing full trajectories. The resulting pipeline improves quality metrics over the prior four-step chunk-wise baseline while cutting latency and training cost. Readers would care because the change directly supports lower-latency streaming and controllable rollout.

Core claim

Causal consistency distillation learns the same AR-conditional flow map as causal ODE distillation yet obtains usable supervision from only a single online teacher ODE step between adjacent timesteps, removing the need to precompute and store full PF-ODE trajectories; this initialization makes the overall Causal Forcing++ pipeline both more efficient and easier to optimize, allowing it to surpass the prior state-of-the-art four-step chunk-wise Causal Forcing under the frame-wise two-step setting.

What carries the argument

Causal consistency distillation, which matches the AR-conditional flow map by supervising with one online teacher ODE step between adjacent timesteps.

If this is right

  • Frame-wise two-step generation exceeds the prior four-step chunk-wise baseline by 0.1 on VBench Total, 0.3 on VBench Quality, and 0.335 on VisionReward.
  • First-frame latency drops by 50 percent.
  • Stage 2 training cost falls by a factor of approximately four.
  • The same pipeline extends to action-conditioned world-model generation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The single-step supervision pattern could be tested on autoregressive models in other modalities where precomputing full trajectories is expensive.
  • Further reduction below two steps per frame might be checked by tightening the consistency loss schedule while keeping the online teacher step fixed.
  • Real-time control loops could be closed around the lower-latency rollout to measure stability in interactive settings.

Load-bearing premise

Causal consistency distillation produces the identical AR-conditional flow map as causal ODE distillation when given supervision from only a single online teacher ODE step.

What would settle it

A side-by-side sampling comparison in which the AR-conditional flow map learned by causal consistency distillation is shown to generate a measurably different distribution of videos than the map learned by causal ODE distillation under identical conditioning.

Figures

Figures reproduced from arXiv: 2605.15141 by Bokai Yan, Chongxuan Li, Hongzhou Zhu, Jun Zhu, Kaiwen Zheng, Min Zhao, Xiao Yang, Xinyuan Li, Zihan Zhou.

Figure 1
Figure 1. Figure 1: Overall framework of our Causal Forcing++ and the comparison with existing methods. (a) Causal Forcing (CF) fixes Self Forcing (SF)’s frame-level injectivity flaw but remains costly; our Causal Forcing++ (CF++) is theoretically sound, efficient, and scalable. (b) Our CF++ reduces training cost by 4×, requires no extra data curation, and achieves 50% lower latency and higher VBench scores than SF and CF. in… view at source ↗
Figure 2
Figure 2. Figure 2: Performance of existing initialization methods after DMD. Self Forcing ODE initialization and multi￾step AR diffusion initialization leads to poor quality and degrade as the setting becomes more aggressive, while Causal Forcing’s causal ODE initialization performs well but is costly and therefore difficult to scale. Multi-step AR diffusion initialization degrades sharply in aggressive settings. Using the m… view at source ↗
Figure 3
Figure 3. Figure 3: Performance and efficiency comparison between causal CD and causal ODE. Causal CD outperforms causal ODE, while subsequently improving training efficiency in time and storage. Causal Forcing++. Putting these pieces together, our pipeline inherits Stage 1 (teacher forcing AR diffu￾sion training) and Stage 3 (asymmetric DMD with self-rollout) from Causal Forcing [20], and replaces its Stage 2 with the causal… view at source ↗
Figure 4
Figure 4. Figure 4: Application of Causal Forcing++ to the action-conditioned world model. Causal Forcing++ enables efficient, high-quality distillation from a bidirectional model to a low-latency AR model, thereby enabling interactive generation toward a Genie3-style world model. score distillation actually works. Causal DMD with teacher forcing. We begin by adapting score distillation into a teacher forcing causal form. Spe… view at source ↗
Figure 5
Figure 5. Figure 5: Performance comparison between causal CD and causal DMD and the intuitive explanation. Causal DMD achieves better early-frame quality than causal CD, but suffers from severe exposure bias in later frames, since mode-seeking DMD is more sensitive to accumulated errors. camera shifts, and eventually degrade to an unacceptable level. In contrast, although causal CD also suffers from gradual quality decay duri… view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison. Our Causal Forcing++ achieves quality and dynamics comparable to or even surpassing Causal Forcing, while outperforming CausVid and Self Forcing. Causal CD is a stronger and more efficient substitute for causal ODE. Causal CD matches or surpasses causal ODE across all step settings while dramatically reducing Stage 2 cost. In 1-step generation, it im￾proves Total, Quality, and Dynam… view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparisons after asymmetric DMD with different initializations. Causal CD achieves results comparable to or even better than, causal ODE, whereas multi-step and causal DMD initializations perform worse. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
read the original abstract

Real-time interactive video generation requires low-latency, streaming, and controllable rollout. Existing autoregressive (AR) diffusion distillation methods have achieved strong results in the chunk-wise 4-step regime by distilling bidirectional base models into few-step AR students, but they remain limited by coarse response granularity and non-negligible sampling latency. In this paper, we study a more aggressive setting: frame-wise autoregression with only 1--2 sampling steps. In this regime, we identify the initialization of a few-step AR student as the key bottleneck: existing strategies are either target-misaligned, incapable of few-step generation, or too costly to scale. We propose \textbf{Causal Forcing++}, a principled and scalable pipeline that uses \emph{causal consistency distillation} (causal CD) for few-step AR initialization. The core idea is that causal CD learns the same AR-conditional flow map as causal ODE distillation, but obtains supervision from a single online teacher ODE step between adjacent timesteps, avoiding the need to precompute and store full PF-ODE trajectories. This makes the initialization both more efficient and easier to optimize. The resulting pipeline, \ours, surpasses the SOTA 4-step chunk-wise Causal Forcing under the \textit{\textbf{frame-wise 2-step setting}} by 0.1 in VBench Total, 0.3 in VBench Quality, and 0.335 in VisionReward, while reducing first-frame latency by 50\% and Stage 2 training cost by $\sim$$4\times$. We further extend the pipeline to action-conditioned world model generation in the spirit of Genie3. Project Page: https://github.com/thu-ml/Causal-Forcing and https://github.com/shengshu-ai/minWM .

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The manuscript introduces Causal Forcing++, a pipeline for few-step autoregressive diffusion distillation in real-time interactive video generation. It proposes causal consistency distillation (causal CD) to initialize AR students in the aggressive frame-wise 1-2 step regime, claiming this learns the identical AR-conditional flow map as causal ODE distillation while obtaining supervision from only a single online teacher ODE step between adjacent timesteps (avoiding full PF-ODE trajectory storage). The resulting method is reported to surpass the SOTA 4-step chunk-wise Causal Forcing baseline under the frame-wise 2-step setting by 0.1 VBench Total, 0.3 VBench Quality, and 0.335 VisionReward, while cutting first-frame latency by 50% and Stage 2 training cost by ~4×; the pipeline is further extended to action-conditioned world models.

Significance. If the core equivalence holds and the metric gains prove robust under controlled ablations, the work would provide a more scalable initialization route for low-latency AR video diffusion, directly addressing the target-misalignment bottleneck in few-step frame-wise regimes and enabling practical streaming generation with reduced compute.

major comments (2)
  1. [Abstract] Abstract: the central claim that causal CD learns the identical AR-conditional flow map as causal ODE distillation via supervision from a single online teacher ODE step lacks any referenced derivation, fixed-point analysis, or section establishing that the resulting student vector field matches the target exactly in the AR setting; without this, the reported gains cannot be attributed to the claimed mechanism rather than initialization artifacts or other factors.
  2. [Abstract] Abstract / Methods: the assertion that the single-step online supervision avoids bias in the frame-wise regime (and thereby enables the 50% latency and 4× cost reductions) is load-bearing for the scalability claim, yet no analysis or equation is provided showing that the approximation preserves the AR-conditional flow map without introducing misalignment.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which identify a need for stronger theoretical grounding of the core claims. We address each point below and will revise the manuscript accordingly to include the requested analysis.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that causal CD learns the identical AR-conditional flow map as causal ODE distillation via supervision from a single online teacher ODE step lacks any referenced derivation, fixed-point analysis, or section establishing that the resulting student vector field matches the target exactly in the AR setting; without this, the reported gains cannot be attributed to the claimed mechanism rather than initialization artifacts or other factors.

    Authors: We agree that an explicit derivation is required to rigorously support the equivalence claim. The manuscript motivates causal CD by construction (causal masking plus single online teacher step between adjacent timesteps), but does not include a fixed-point analysis. In revision we will add a new subsection (e.g., 3.3) containing the derivation: we show that the causal consistency loss yields a student vector field whose fixed point coincides with the AR-conditional probability flow ODE under the same teacher, with the single-step supervision preserving the flow map exactly in the limit. This will allow attribution of gains to the mechanism rather than artifacts. revision: yes

  2. Referee: [Abstract] Abstract / Methods: the assertion that the single-step online supervision avoids bias in the frame-wise regime (and thereby enables the 50% latency and 4× cost reductions) is load-bearing for the scalability claim, yet no analysis or equation is provided showing that the approximation preserves the AR-conditional flow map without introducing misalignment.

    Authors: We concur that an equation-level demonstration is needed to substantiate the bias-avoidance claim. The current text states the practical benefit but omits the supporting analysis. In the revision we will insert the relevant equations showing that the online single-step teacher update between adjacent frames maintains the AR-conditional flow without the misalignment that arises from full-trajectory precomputation or bidirectional approximations; this directly justifies the reported latency and cost reductions under the frame-wise 1-2 step regime. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained against external benchmarks

full rationale

The paper's central claim—that causal consistency distillation learns the identical AR-conditional flow map as causal ODE distillation via single online teacher ODE step—is presented as a core idea without any quoted equations, fitted parameters, or self-citations that reduce the target AR flow map to a definitional identity or prior self-citation chain. No self-definitional steps, fitted-input predictions, or uniqueness theorems imported from overlapping authors appear in the provided text. Performance numbers are framed as empirical outcomes of the pipeline rather than quantities forced by construction. The derivation chain therefore remains independent and falsifiable outside the paper's own fitted values.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5887 in / 1067 out tokens · 28385 ms · 2026-06-30T20:57:25.158917+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Causal-rCM: A Unified Teacher-Forcing and Self-Forcing Open Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation and Interactive World Models

    cs.CV 2026-06 unverdicted novelty 6.0

    Causal-rCM unifies teacher-forcing and self-forcing distillation for autoregressive video diffusion, delivering a 2-step model with VBench-T2V score 84.63 and enabling interactive world models on Cosmos 3 using only s...

  2. minWM: A Full-Stack Open-Source Framework for Real-Time Interactive Video World Models

    cs.CV 2026-05 unverdicted novelty 6.0

    minWM supplies an end-to-end pipeline that fine-tunes bidirectional T2V/TI2V models with camera control then distills them via Causal Forcing into few-step autoregressive generators for low-latency rollout.

  3. One-Forcing: Towards Stable One-Step Autoregressive Video Generation

    cs.CV 2026-05 unverdicted novelty 5.0

    One-Forcing augments DMD with a GAN loss to enable stable one-step causal autoregressive video generation, reporting a VBench score of 83.76 as SOTA among one-step methods.

Reference graph

Works this paper leans on

67 extracted references · 46 canonical work pages · cited by 3 Pith papers · 29 internal anchors

  1. [1]

    Video generation models as world simulators

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr , Joe T aylor , Troy Luhman, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. Video generation models as world simulators. 2024

    2024

  2. [2]

    Vidu: a highly consistent, dynamic and skilled text-to-video generator with diffusion models.arXiv preprint arXiv:2405.04233, 2024

    Fan Bao, Chendong Xiang, Gang Yue, Guande He, Hongzhou Zhu, Kaiwen Zheng, Min Zhao, Shilong Liu, Y aole Wang, and Jun Zhu. Vidu: a highly consistent, dynamic and skilled text-to-video generator with diffusion models. arXiv preprint arXiv:2405.04233 , 2024

    work page arXiv 2024

  3. [3]

    Wan: Open and Advanced Large-Scale Video Generative Models

    T eam Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Y ang, et al. Wan: Open and advanced large-scale video generative models. arXiv preprint arXiv:2503.20314 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  4. [4]

    HunyuanVideo: A Systematic Framework For Large Video Generative Models

    Weijie Kong, Qi Tian, Zijian Zhang, Rox Min, Zuozhuo Dai, Jin Zhou, Jiangfeng Xiong, Xin Li, Bo Wu, Jian- wei Zhang, et al. Hunyuanvideo: A systematic framework for large video generative models. arXiv preprint arXiv:2412.03603, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  5. [5]

    CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

    Zhuoyi Y ang, Jiayan T eng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Y ang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al. Cogvideox: T ext-to-video diffusion models with an expert transformer . arXiv preprint arXiv:2408.06072, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  6. [6]

    Open-Sora Plan: Open-Source Large Video Generation Model

    Bin Lin, Yunyang Ge, Xinhua Cheng, Zongjian Li, Bin Zhu, Shaodong Wang, Xianyi He, Y ang Y e, Shenghai Yuan, Liuhan Chen, et al. Open-sora plan: Open-source large video generation model. arXiv preprint arXiv:2412.00131 , 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  7. [7]

    Open-Sora: Democratizing Efficient Video Production for All

    Zangwei Zheng, Xiangyu Peng, Tianji Y ang, Chenhui Shen, Shenggui Li, Hongxin Liu, Yukun Zhou, Tianyi Li, and Y ang Y ou. Open-sora: Democratizing efficient video production for all. arXiv preprint arXiv:2412.20404 , 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modeling

    Wenqiang Sun, Haiyu Zhang, Haoyuan Wang, Junta Wu, Zehan Wang, Zhenwei Wang, Yunhong Wang, Jun Zhang, T engfei Wang, and Chunchao Guo. Worldplay: T owards long-term geometric consistency for real-time interactive world modeling. arXiv preprint arXiv:2512.14614 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  9. [9]

    Philip J. Ball, Jakob Bauer , Frank Belletti, Bethanie Brownfield, Ariel Ephrat, Shlomi Fruchter , Agrim Gupta, Kris- tian Holsheimer , Aleksander Holynski, Jiri Hron, Christos Kaplanis, Marjorie Limont, Matt McGill, Y anko Oliveira, Jack Parker-Holder , Frank Perbet, Guy Scully , Jeremy Shar , Stephen Spencer , Omer T ov , Ruben Villegas, Emma Wang, Jessi...

    2025

  10. [10]

    Live Avatar: Streaming Real-time Audio-Driven Avatar Generation with Infinite Length

    Yubo Huang, Hailong Guo, Fangtai Wu, Shifeng Zhang, Shijie Huang, Qijun Gan, Lin Liu, Sirui Zhao, Enhong Chen, Jiaming Liu, et al. Live avatar: Streaming real-time audio-driven avatar generation with infinite length. arXiv preprint arXiv:2512.04677, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    Avatar Forcing: Real-Time Interactive Head Avatar Generation for Natural Conversation

    T aekyung Ki, Sangwon Jang, Jaehyeong Jo, Jaehong Y oon, and Sung Ju Hwang. Avatar forcing: Real-time interac- tive head avatar generation for natural conversation. arXiv preprint arXiv:2601.00664 , 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  12. [12]

    Streamavatar: Streaming diffusion models for real-time interactive human avatars.arXiv preprint arXiv:2512.22065, 2025

    Zhiyao Sun, Ziqiao Peng, Yifeng Ma, Yi Chen, Zhengguang Zhou, Zixiang Zhou, Guozhen Zhang, Y ouliang Zhang, Yuan Zhou, Qinglin Lu, et al. Streamavatar: Streaming diffusion models for real-time interactive human avatars. arXiv preprint arXiv:2512.22065 , 2025

    work page arXiv 2025

  13. [13]

    Vidarc: Embodied video diffusion model for closed-loop control.arXiv preprint arXiv:2512.17661, 2025

    Y ao Feng, Chendong Xiang, Xinyi Mao, Hengkai T an, Zuyue Zhang, Shuhe Huang, Kaiwen Zheng, Haitian Liu, Hang Su, and Jun Zhu. Vidarc: Embodied video diffusion model for closed-loop control. arXiv preprint arXiv:2512.17661, 2025

    work page arXiv 2025

  14. [14]

    World action models are zero-shot policies, 2026

    Seonghyeon Y e, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, Y ou Liang T an, Chuning Zhu, Jiannan Xiang, Ayaan Malik, Kyungmin Lee, William Liang, Nadun Ranawaka, Jiasheng Gu, Yinzhen Xu, Guanzhi Wang, Fengyuan Hu, Avnish Narayan, Johan Bjorck, Jing Wang, Gwanghyun Kim, Dantong Niu, Ruijie Zheng, Yuqi Xie, Jimmy Wu, ...

    2026

  15. [15]

    Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954, 2024

    Y ang Jin, Zhicheng Sun, Ningyuan Li, Kun Xu, Hao Jiang, Nan Zhuang, Quzhe Huang, Y ang Song, Y adong Mu, and Zhouchen Lin. Pyramidal flow matching for efficient video generative modeling. arXiv preprint arXiv:2410.05954 , 2024

    work page arXiv 2024

  16. [16]

    MAGI-1: Autoregressive Video Generation at Scale

    Hansi T eng, Hongyu Jia, Lei Sun, Lingzhi Li, Maolin Li, Mingqiu T ang, Shuai Han, Tianning Zhang, WQ Zhang, Weifeng Luo, et al. Magi-1: Autoregressive video generation at scale. arXiv preprint arXiv:2505.13211 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  17. [17]

    SkyReels-V2: Infinite-length Film Generative Model

    Guibin Chen, Dixuan Lin, Jiangping Y ang, Chunze Lin, Junchen Zhu, Mingyuan Fan, Hao Zhang, Sheng Chen, Zheng Chen, Chengcheng Ma, et al. Skyreels-v2: Infinite-length film generative model. arXiv preprint arXiv:2504.13074, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  18. [18]

    From slow bidirectional to fast autoregressive video diffusion models

    Tianwei Yin, Qiang Zhang, Richard Zhang, William T Freeman, Fredo Durand, Eli Shechtman, and Xun Huang. From slow bidirectional to fast autoregressive video diffusion models. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 22963–22974, 2025

    2025

  19. [19]

    Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusion

    Xun Huang, Zhengqi Li, Guande He, Mingyuan Zhou, and Eli Shechtman. Self forcing: Bridging the train-test gap in autoregressive video diffusion. arXiv preprint arXiv:2506.08009 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  20. [20]

    Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation

    Hongzhou Zhu, Min Zhao, Guande He, Hang Su, Chongxuan Li, and Jun Zhu. Causal forcing: Autoregressive diffu- sion distillation done right for high-quality real-time interactive video generation. arXiv preprint arXiv:2602.02214, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  21. [21]

    Consistency models

    Y ang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever . Consistency models. 2023

    2023

  22. [22]

    Advances in neural information processing systems37, 83951–84009 (2024),https://arxiv.org/abs/ 2405.18407

    Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman, Dazhong Shen, Peng Gao, Michael Lingelbach, Ke- qiang Sun, Weikang Bian, Guanglu Song, Yu Liu, et al. Phased consistency model. arXiv preprint arXiv:2405.18407, 2024

    work page arXiv 2024

  23. [23]

    Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models

    Cheng Lu and Y ang Song. Simplifying, stabilizing and scaling continuous-time consistency models. arXiv preprint arXiv:2410.11081, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  24. [24]

    Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency

    Kaiwen Zheng, Yuji Wang, Qianli Ma, Huayu Chen, Jintao Zhang, Y ogesh Balaji, Jianfei Chen, Ming-Yu Liu, Jun Zhu, and Qinsheng Zhang. Large scale diffusion distillation via score-regularized continuous-time consistency . arXiv preprint arXiv:2510.08431 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  25. [25]

    Instaflow: One step is enough for high-quality diffusion- based text-to-image generation

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffusion- based text-to-image generation. In The Twelfth International Conference on Learning Representations , 2023

    2023

  26. [26]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  27. [27]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Y ang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar , Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 , 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  28. [28]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow . arXiv preprint arXiv:2209.03003 , 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  29. [29]

    Flow Matching for Generative Modeling

    Y aron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. arXiv preprint arXiv:2210.02747 , 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  30. [30]

    Egsde: Unpaired image-to-image translation via energy-guided stochastic differential equations

    Min Zhao, Fan Bao, Chongxuan Li, and Jun Zhu. Egsde: Unpaired image-to-image translation via energy-guided stochastic differential equations. Advances in Neural Information Processing Systems , 35:3609–3623, 2022

    2022

  31. [31]

    Identifying and solving conditional image leakage in image-to-video diffusion model

    Min Zhao, Hongzhou Zhu, Chendong Xiang, Kaiwen Zheng, Chongxuan Li, and Jun Zhu. Identifying and solving conditional image leakage in image-to-video diffusion model. Advances in Neural Information Processing Systems , 37:30300–30326, 2024

    2024

  32. [32]

    Controlvideo: conditional control for one-shot text-driven video editing and beyond

    Min Zhao, Rongzhen Wang, Fan Bao, Chongxuan Li, and Jun Zhu. Controlvideo: conditional control for one-shot text-driven video editing and beyond. Science China Information Sciences , 68(3):132107, 2025

    2025

  33. [33]

    arXiv preprint arXiv:2502.15894 (2025)

    Min Zhao, Guande He, Yixiao Chen, Hongzhou Zhu, Chongxuan Li, and Jun Zhu. Riflex: A free lunch for length extrapolation in video diffusion transformers. arXiv preprint arXiv:2502.15894 , 2025. 16 T echnical Report

    work page arXiv 2025

  34. [34]

    Ultravico: Breaking extrapolation limits in video diffusion transformers

    Min Zhao, Hongzhou Zhu, Yingze Wang, Bokai Y an, Jintao Zhang, Guande He, Ling Y ang, Chongxuan Li, and Jun Zhu. Ultravico: Breaking extrapolation limits in video diffusion transformers. arXiv preprint arXiv:2511.20123 , 2025

    work page arXiv 2025

  35. [35]

    Ul- traimage: Rethinking resolution extrapolation in image diffusion transformers

    Min Zhao, Bokai Y an, Xue Y ang, Hongzhou Zhu, Jintao Zhang, Shilong Liu, Chongxuan Li, and Jun Zhu. Ul- traimage: Rethinking resolution extrapolation in image diffusion transformers. arXiv preprint arXiv:2512.04504 , 2025

    work page arXiv 2025

  36. [36]

    Radial attention:𝒪(𝑛log𝑛)sparse attention with energy decay for long video generation.arXiv preprint arXiv:2506.19852, 2025

    Xingyang Li, Muyang Li, Tianle Cai, Haocheng Xi, Shuo Y ang, Yujun Lin, Lvmin Zhang, Songlin Y ang, Jinbo Hu, Kelly Peng, et al. Radial attention: o(n log n) sparse attention with energy decay for long video generation. arXiv preprint arXiv:2506.19852, 2025

    work page arXiv 2025

  37. [37]

    arXiv preprint arXiv:2104.14806 , year=

    Chenfei Wu, Lun Huang, Qianxi Zhang, Binyang Li, Lei Ji, Fan Y ang, Guillermo Sapiro, and Nan Duan. Godiva: Generating open-domain videos from natural descriptions. arXiv preprint arXiv:2104.14806 , 2021

    work page arXiv 2021

  38. [38]

    CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers

    Wenyi Hong, Ming Ding, Wendi Zheng, Xinghan Liu, and Jie T ang. Cogvideo: Large-scale pretraining for text-to- video generation via transformers. arXiv preprint arXiv:2205.15868 , 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  39. [39]

    Nüwa: Visual synthesis pre-training for neural visual world creation

    Chenfei Wu, Jian Liang, Lei Ji, Fan Y ang, Yuejian Fang, Daxin Jiang, and Nan Duan. Nüwa: Visual synthesis pre-training for neural visual world creation. In European conference on computer vision , pages 720–736. Springer , 2022

    2022

  40. [40]

    Weissenborn, O

    Dirk Weissenborn, Oscar Täckström, and Jakob Uszkoreit. Scaling autoregressive video models. arXiv preprint arXiv:1906.02634, 2019

    work page arXiv 1906

  41. [41]

    VideoGPT: Video Generation using VQ-VAE and Transformers

    Wilson Y an, Yunzhi Zhang, Pieter Abbeel, and Aravind Srinivas. Videogpt: Video generation using vq-vae and transformers. arXiv preprint arXiv:2104.10157 , 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  42. [42]

    Autoregressive Video Generation without Vector Quantization

    Haoge Deng, Ting Pan, Haiwen Diao, Zhengxiong Luo, Yufeng Cui, Huchuan Lu, Shiguang Shan, Y onggang Qi, and Xinlong Wang. Autoregressive video generation without vector quantization. arXiv preprint arXiv:2412.14169 , 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  43. [43]

    VideoPoet: A Large Language Model for Zero-Shot Video Generation

    Dan Kondratyuk, Lijun Yu, Xiuye Gu, José Lezama, Jonathan Huang, Grant Schindler , Rachel Hornung, Vighnesh Birodkar , Jimmy Y an, Ming-Chang Chiu, et al. Videopoet: A large language model for zero-shot video generation. arXiv preprint arXiv:2312.14125 , 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  44. [44]

    Diffusion forcing: Next-token prediction meets full-sequence diffusion

    Boyuan Chen, Diego Martí Monsó, Yilun Du, Max Simchowitz, Russ T edrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffusion. Advances in Neural Information Processing Systems , 37:24081–24125, 2024

    2024

  45. [45]

    Pack and force your memory: Long-form and consistent video generation.arXiv preprint arXiv:2510.01784, 2025

    Xiaofei Wu, Guozhen Zhang, Zhiyong Xu, Yuan Zhou, Qinglin Lu, and Xuming He. Pack and force your memory: Long-form and consistent video generation. arXiv preprint arXiv:2510.01784 , 2025

    work page arXiv 2025

  46. [46]

    End-to-End Training for Autoregressive Video Diffusion via Self-Resampling

    Yuwei Guo, Ceyuan Y ang, Hao He, Y ang Zhao, Meng Wei, Zhenheng Y ang, Weilin Huang, and Dahua Lin. End-to- end training for autoregressive video diffusion via self-resampling. arXiv preprint arXiv:2512.15702 , 2025

    work page internal anchor Pith review arXiv 2025

  47. [47]

    Bagger: Backwards aggregation for mitigating drift in autoregressive video diffusion models.arXiv preprint arXiv:2512.12080, 2025

    Ryan Po, Eric Ryan Chan, Changan Chen, and Gordon Wetzstein. Bagger: Backwards aggregation for mitigating drift in autoregressive video diffusion models. arXiv preprint arXiv:2512.12080 , 2025

    work page arXiv 2025

  48. [48]

    Autoregressive adversarial post-training for real-time interactive video generation.arXiv preprint arXiv:2506.09350, 2025

    Shanchuan Lin, Ceyuan Y ang, Hao He, Jianwen Jiang, Yuxi Ren, Xin Xia, Y ang Zhao, Xuefeng Xiao, and Lu Jiang. Autoregressive adversarial post-training for real-time interactive video generation. arXiv preprint arXiv:2506.09350, 2025

    work page arXiv 2025

  49. [49]

    Diffusion adversarial post-training for one-step video generation.arXiv preprint arXiv:2501.08316,

    Shanchuan Lin, Xin Xia, Yuxi Ren, Ceyuan Y ang, Xuefeng Xiao, and Lu Jiang. Diffusion adversarial post-training for one-step video generation. arXiv preprint arXiv:2501.08316 , 2025

    work page arXiv 2025

  50. [50]

    T owards one-step causal video generation via adversarial self-distillation

    Y ongqi Y ang, Huayang Huang, Xu Peng, Xiaobin Hu, Donghao Luo, Jiangning Zhang, Chengjie Wang, and Yu Wu. T owards one-step causal video generation via adversarial self-distillation. arXiv preprint arXiv:2511.01419 , 2025

    work page arXiv 2025

  51. [51]

    Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillation

    Yunhong Lu, Y anhong Zeng, Haobo Li, Hao Ouyang, Qiuyu Wang, Ka Leong Cheng, Jiapeng Zhu, Hengyuan Cao, Zhipeng Zhang, Xing Zhu, et al. Reward forcing: Efficient streaming video generation with rewarded distribution matching distillation. arXiv preprint arXiv:2512.04678 , 2025. 17 T echnical Report

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    Worldcompass: Reinforce- ment learning for long-horizon world models,

    Zehan Wang, T engfei Wang, Haiyu Zhang, Xuhui Zuo, Junta Wu, Haoyuan Wang, Wenqiang Sun, Zhenwei Wang, Chenjie Cao, Hengshuang Zhao, et al. Worldcompass: Reinforcement learning for long-horizon world models. arXiv preprint arXiv:2602.09022 , 2026

    work page arXiv 2026

  53. [53]

    LongLive: Real-time Interactive Long Video Generation

    Shuai Y ang, Wei Huang, Ruihang Chu, Yicheng Xiao, Yuyang Zhao, Xianbang Wang, Muyang Li, Enze Xie, Yingcong Chen, Y ao Lu, et al. Longlive: Real-time interactive long video generation. arXiv preprint arXiv:2509.22622 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  54. [54]

    Rolling Forcing: Autoregressive Long Video Diffusion in Real Time

    Kunhao Liu, Wenbo Hu, Jiale Xu, Ying Shan, and Shijian Lu. Rolling forcing: Autoregressive long video diffusion in real time. arXiv preprint arXiv:2509.25161 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  55. [55]

    Self-Forcing++: Towards Minute-Scale High-Quality Video Generation

    Justin Cui, Jie Wu, Ming Li, T ao Y ang, Xiaojie Li, Rui Wang, Andrew Bai, Yuanhao Ban, and Cho-Jui Hsieh. Self- forcing++: T owards minute-scale high-quality video generation. arXiv preprint arXiv:2510.02283 , 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  56. [56]

    Prolificdreamer: High- fidelity and diverse text-to-3d generation with variational score distillation

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High- fidelity and diverse text-to-3d generation with variational score distillation. Advances in neural information process- ing systems, 36:8406–8441, 2023

    2023

  57. [57]

    Diff-instruct: A universal approach for transferring knowledge from pre-trained diffusion models

    Weijian Luo, Tianyang Hu, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, and Zhihua Zhang. Diff-instruct: A universal approach for transferring knowledge from pre-trained diffusion models. Advances in Neural Information Processing Systems, 36:76525–76546, 2023

    2023

  58. [58]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T Freeman, and T aesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 6613–6623, 2024

    2024

  59. [59]

    Improved Techniques for Training Consistency Models

    Y ang Song and Prafulla Dhariwal. Improved techniques for training consistency models. arXiv preprint arXiv:2310.14189, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  60. [60]

    Vbench: Comprehensive benchmark suite for video generative models

    Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianxing Wu, Qingyang Jin, Nattapol Chanpaisit, et al. Vbench: Comprehensive benchmark suite for video generative models. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 21807–21818, 2024

    2024

  61. [61]

    Cameras as relative positional encoding

    Ruilong Li, Brent Yi, Junchen Liu, Hang Gao, Yi Ma, and Angjoo Kanazawa. Cameras as relative positional encoding. In D. Belgrave, C. Zhang, H. Lin, R. Pascanu, P. Koniusz, M. Ghassemi, and N. Chen, editors, Advances in Neural Information Processing Systems , volume 38, pages 15984–16009. Curran Associates, Inc., 2025

    2025

  62. [62]

    Improved distribution matching distillation for fast image synthesis

    Tianwei Yin, Michaël Gharbi, T aesung Park, Richard Zhang, Eli Shechtman, Fredo Durand, and William T Freeman. Improved distribution matching distillation for fast image synthesis. In NeurIPS, 2024

    2024

  63. [63]

    Autoregressive adversarial post-training for real-time interactive video generation

    Shanchuan Lin, Ceyuan Y ang, Hao He, Jianwen Jiang, Yuxi Ren, Xin Xia, Y ang Zhao, Xuefeng Xiao, and Lu Jiang. Autoregressive adversarial post-training for real-time interactive video generation. Advances in Neural Information Processing Systems, 38:41061–41086, 2026

    2026

  64. [64]

    Generative adversarial nets

    Ian J Goodfellow , Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley , Sherjil Ozair , Aaron Courville, and Y oshua Bengio. Generative adversarial nets. Advances in neural information processing systems , 27, 2014

    2014

  65. [65]

    OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation

    Kepan Nan, Rui Xie, Penghao Zhou, Tiehan Fan, Zhenheng Y ang, Zhijie Chen, Xiang Li, Jian Y ang, and Ying T ai. Openvid-1m: A large-scale high-quality dataset for text-to-video generation. arXiv preprint arXiv:2407.02371 , 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  66. [66]

    Vidprom: A million-scale real prompt-gallery dataset for text-to-video diffusion models

    Wenhao Wang and Yi Y ang. Vidprom: A million-scale real prompt-gallery dataset for text-to-video diffusion models. Advances in Neural Information Processing Systems , 37:65618–65642, 2024

    2024

  67. [67]

    VisionReward: Fine-Grained Multi-Dimensional Human Preference Learning for Image and Video Generation

    Jiazheng Xu, Yu Huang, Jiale Cheng, Yuanming Y ang, Jiajun Xu, Yuan Wang, Wenbo Duan, Shen Y ang, Qunlin Jin, Shurun Li, et al. Visionreward: Fine-grained multi-dimensional human preference learning for image and video generation. arXiv preprint arXiv:2412.21059 , 2024. 18

    work page internal anchor Pith review Pith/arXiv arXiv 2024