Pith. sign in

REVIEW 5 major objections 6 minor 3 cited by

Transition Matching Distillation for Fast Video Generation

T0 review · 5 major / 6 minor · reviewed 2026-08-03 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A new distillation framework, Transition Matching Distillation, compresses pretrained video diffusion models into one-to-few-step generators whose output distribution matches the teacher's, beating prior distilled models at the same inferen

desk verdict Decoupled-backbone distillation is a real step forward for fast video generation, but the headline 'outperforms existing models' is undercut by a 14B-generated training-data confound and the admitted 14B two-step shortfall. read the letter →

arxiv 2601.09881 v2 pith:RXETR5JA submitted 2026-01-14 cs.CV cs.AIcs.LG

Weili Nie , Julius Berner , Nanye Ma , Chao Liu , Saining Xie , Arash Vahdat This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords videogenerationdiffusiondistillationfew-stepsamplingtransitionmatchingflowdistributionDMD2Wan2.1
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to speed up large video diffusion models, which normally require dozens of denoising steps, by distilling them into generators that take one or two effective forward passes. Its central object is a decoupled architecture: a main backbone that computes semantic features once per outer step, and a lightweight recurrent flow head that performs several inner refinements conditioned on those features. The training is a two-stage process — a transition-matching pretraining that turns the head into a conditional flow map, followed by a distribution-matching distillation that aligns the student's transition process with the teacher's denoising distribution. The authors report that the resulting model outperforms existing distilled video generators on the benchmarks they use, and that a near-one-step 14B student reaches an overall score of 84.24 on a standard video benchmark. The work matters because it suggests that high-quality video generation can become fast enough for interactive use without sacrificing the teacher's semantic fidelity.

What carries the argument

The central objects are the decoupled student architecture and the transition process. The main backbone (early layers of the pretrained DiT) extracts a semantic feature m_t for each outer step; the flow head (the last few layers, re-initialized) performs N inner flow updates conditioned on m_t via a gated fusion layer, with the DTM target y = x1 - x. The head is pretrained with a MeanFlow objective (Eq. 9-10) to learn a conditional inner flow map, and then the whole unrolled generator is trained with a video-adapted DMD2 objective (VSD + GAN) that backpropagates through all inner steps. This combination is what allows the few-step transition to approximate the teacher's many-step trajectory

What would settle it

Independently reproduce the two-step distillation of the 1.3B model with the published recipe and a separate dataset, and check whether the overall score on the same benchmark stays around 84.7. Alternatively, measure the student's output distribution against the teacher's on a wide prompt set (e.g., using a distributional distance) to see whether the match is genuine rather than benchmark-specific.

Watch

Extended reading notes

Core claim

The discovery is that a many-step denoising trajectory can be replaced by a compact probability transition process without losing the teacher's distribution, provided the student is split into a semantic backbone and a flow head. Each outer transition step unrolls the flow head for N inner steps to refine fine details, and the whole unrolled path is trained end-to-end with a distribution-matching objective. The paper demonstrates this on two Wan2.1 text-to-video models, showing that effective NFE can be reduced to 1.38 for a 14B model while improving over one-step baselines and remaining at parity with or above prior distilled models at corresponding costs. The authors attribute the gains to

Load-bearing premise

The premise that the early layers of the pretrained model, evaluated once per outer step, provide a sufficient condition for the inner flow's velocity, so that the last few layers can be re-initialized as a head without the student's distribution collapsing; if this decoupling is invalid, the two-stage training cannot recover the teacher's distribution.

Editorial extensions

If this is right

  • Distilled video models can reach one-to-two-step generation with quality at or above prior distilled models at the same effective NFE, enabling real-time or interactive video synthesis.
  • The decoupled design offers fine-grained control over the speed/quality tradeoff via the number of inner steps N and flow-head layers H, including fractional effective NFE.
  • The ablation results imply that flow-head rollout during distillation is necessary for the gains, and that MeanFlow pretraining is superior to vanilla flow-matching pretraining as a first stage.
  • The improved DMD2 recipe (Conv3D discriminator, timestep shifting, selective KD warm-up) is itself a stronger baseline for video distillation.
  • The method reports a 14B one-step model at 84.24 overall score on a standard benchmark, closing much of the gap to the 50-step teacher at 86.22.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The decoupled backbone/head structure might transfer to other large generative models (e.g., autoregressive video models), not just flow-based diffusion teachers, since the split only assumes that early-layer features are a sufficient conditioning signal.
  • The inner-flow rollout could be viewed as a lightweight 'refinement head' that might be trained with less teacher supervision, e.g., using self-supervised or synthetic targets, which the paper does not explore.
  • If the reported benchmark scores are confirmed by independent replication, near-real-time video generation on consumer hardware becomes plausible when combined with the system-level optimizations the paper lists as orthogonal (efficient attention, feature caching).
  • The method's dependence on a 500k synthetic text-video dataset for distillation raises the question of whether gains persist when the teacher is smaller or the data distribution shifts; this is untested in the paper.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes Transition Matching Distillation (TMD), a two-stage framework for distilling pretrained video flow models into few-step generators. The student network decouples the teacher's DiT backbone into a semantic main backbone and a lightweight flow head; Stage 1 uses a MeanFlow-style transition matching objective to train the flow head as a conditional inner-flow map, and Stage 2 applies an improved DMD2 objective (DMD2-v) with the flow head unrolled for multiple inner refinement steps. Experiments on Wan2.1 1.3B and 14B report VBench scores and a user preference study, with the headline claim that TMD outperforms existing distilled models at comparable inference cost.

Significance. If the empirical claim is established, the method is a useful step for few-step video generation: the decoupled architecture is simple, the two-stage procedure is clearly specified, and the idea of unrolling a lightweight flow head inside a distribution-matching distillation is well motivated. The paper is commendably detailed: it provides pseudocode for training and inference, full hyperparameters, finite-difference handling of the MeanFlow JVP, ablations of each design choice, and a blinded user study. These are real strengths. However, the central comparative claim is currently under-supported because the main tables mix teachers, the distillation data is generated by a different model, and no uncertainty quantification is supplied. The method itself appears defensible, but the evidence does not yet justify the unqualified statement that TMD outperforms existing distilled models.

major comments (5)
  1. [Section 4.1 and Table 7] The training corpus used for both the 1.3B and 14B distillations consists of 500k videos generated by the Wan2.1 14B model (Table 7). For the 1.3B experiments, the student is therefore trained on data produced by a stronger model than its teacher, while external baselines such as rCM are not trained on this corpus. The headline comparison in Table 1 (TMD-N2H5 84.68 vs rCM-4 84.43) is thus not a controlled comparison of distillation methods; the gain could come from the 14B-generated training distribution. Please add same-data, same-teacher baselines (e.g., rCM and DMD2-v trained on the identical generated corpus) or otherwise control for this confound.
  2. [Table 1 footnotes and Section 4.2] Several rows in Table 1 use different teacher models: DOLLAR uses a CogVideoX variant, T2V-Turbo-v2 uses VideoCrafter2, and APT uses Seaweed-7B. Since these are different base generators, their VBench scores cannot be used to conclude that TMD is a better distillation method. The statement in Section 4.2 that TMD 'outperforms all other distilled models' is therefore too broad. The valid comparison is against methods that distill the same Wan2.1 1.3B teacher, and the tables should be reorganized to make that distinction explicit.
  3. [Table 2 and Abstract] The abstract claims TMD 'outperforms existing distilled models under comparable inference costs,' but Table 2 shows the two-step 14B model TMD-N4H5 (effective NFE=2.75, overall 84.62) is below rCM at NFE=2 (overall 85.05) and DMD2-v at NFE=2 (84.79), despite a higher effective NFE. The paper acknowledges this in Section 4.2 ('does not outperform 2-step baselines') but the abstract and introduction state the claim without this qualification. Please restrict the claim to the configurations where it is actually observed, or provide a convincing explanation for the 14B two-step reversal.
  4. [Tables 1-4 and Figure 5] All VBench comparisons are reported as single point estimates with no variance, confidence intervals, or significance tests. The differences underlying the key claims are small (e.g., Table 1: 84.68 vs 84.43; Table 2: 84.24 vs 83.02), and the user study in Figure 5 reports win rates without confidence intervals or sample sizes per item. Since VBench evaluation involves stochastic video generation, at least a few independently seeded runs for the main configurations and the strongest same-teacher baselines are needed to establish that the gaps are not noise.
  5. [Section 3 and Appendix A.1] The method rests on the assumption that the early-layer features m_theta(x_ti,ti) are a sufficient condition for the inner velocity, and that the gated fusion 'ensures the student's initial forward pass matches that of the teacher.' This is plausible but not directly demonstrated. A concrete diagnostic would help: after Stage 1, compare the student's main-backbone features and flow-head outputs against the corresponding teacher features on identical noisy inputs, or ablate the gating initialization explicitly. Without such a check, the reader cannot tell whether the two-stage procedure is preserving the teacher's trajectory or merely fitting a flexible head to a related but different denoising task.
minor comments (6)
  1. [Section 4.1, Eq. (16)] The effective NFE formula counts N-1 flow-head passes as extra beyond the first pass. Please clarify in the text why the first inner flow step is not counted as additional cost; as written, Algorithm 1 appears to execute N flow-head forward passes per outer step.
  2. [Section 4.2] 'Significantly outperforms' is used for VBench score differences without statistical tests. Please use 'reports a higher score' or add error bars.
  3. [Figure 5] The user study would benefit from reporting the number of raters, the number of pairwise comparisons per condition, and confidence intervals or a significance test.
  4. [Section 3.2 and Appendix B] 'DMD-v' appears in a few places where 'DMD2-v' is meant. Please unify the notation.
  5. [Appendix B.4] The alternative parametrization for y := x appears to have a typo or missing parentheses in the denominator. Please check the formula.
  6. [Related Work] The comparison with Transition Matching could be sharpened: the paper positions TMD as an extension of TM, but it would help to state explicitly which TM design choices are inherited and which are changed for video distillation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: TMD's distillation objectives are standard bootstrapping losses and the main empirical claims are tested against external benchmarks.

full rationale

The core derivation (Section 3, Eqs. 9–15, Algorithm 2) uses two standard training objectives: the MeanFlow self-consistency loss (Eq. 9) and the DMD2 distribution-matching loss (Eq. 11). In both cases the student network is optimized against targets that involve stop-gradients of the same network or teacher/fake-score networks, but no reported result is defined in terms of a fitted parameter that is later renamed as a prediction. The unrolled student output in Eq. (15), g_theta(x_ti, ti; y1) = x1 - InnerFlow(m_theta(x_ti, ti)), is the object being trained, not a fitted input; its quality is evaluated on VBench, an external benchmark. The gated fusion initialization that 'ensures the student's initial forward pass matches that of the teacher' is an architectural initialization choice (Appendix A.1), not a derivation that presupposes the target performance. The comparison against rCM, DOLLAR, T2V-Turbo-v2, and APT uses published numbers on the same VBench protocol; the only notable concern is that the 1.3B distillation used videos generated by the 14B model as training data (Table 7), which is a potential experimental confound for cross-method comparisons, but it is not a circularity because the method's loss does not define its output as that training data. Self-citations in the references (e.g., [26], [54], [68]) are not load-bearing: TMD is built on externally published MeanFlow [17] and DMD2 [58], and no uniqueness theorem or prior author-derived ansatz is invoked to force the architecture or training objective. Therefore, no step in the claimed derivation reduces by construction to its own input.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

Central claim rests mostly on the decoupling assumption and standard distillation losses; no new physical entities are introduced. The load-bearing hand-chosen quantities are timestep shifts and the JVP finite-difference step, which are tuned per model.

free parameters (8)
  • Timestep shift gamma for t_student (outer denoising grid) = 10 (default; 5 for 4-step DMD2-v)
    Chosen by hand; controls concentration of outer transition steps near data. Table 5 shows removing it lowers 2-step overall from 84.39 to 83.44.
  • Timestep shift gamma for t_dmd (noise level in DMD/VSD loss) = 5
    Chosen to avoid mode collapse; Table 5 and Figure 9 show w/o shift causes severe mode collapse not captured by VBench.
  • Timestep shift gamma for s_student (inner flow time grid in TM-MF) = 10
    Hand-picked in Appendix A.2 to align inner flow with outer transition schedule.
  • Timestep shift gamma for s_mf (MeanFlow timestep sampling) = 3
    Hand-picked for TM-MF stabilization, as described in Appendix A.2.
  • Finite-difference step delta for JVP approximation = 0.005
    Chosen empirically in Appendix A.2; replaces exact forward-mode JVP for FSDP/flash-attention compatibility.
  • Loss normalization constant c in TM-MF adaptive loss = d (1.3B), d/1e5 (14B)
    Per-model constants in adaptive loss normalization; d is the dimension of y_s.
  • GAN discriminator loss weight lambda = 0.03
    DMD2-v hyperparameter; balances VSD and adversarial loss.
  • CFG scales = 3 (TM-MF), 5 (DMD2-v)
    Classifier-free guidance weights chosen during distillation and pretraining.
assumptions (5)
  • domain assumption The teacher Wan2.1 model is a pretrained flow/velocity model that satisfies Eq. (2) and can be evaluated with CFG.
    The whole method assumes the teacher's velocity predictions are reliable and the decoupled student is initialized from it (Section 3, Appendix A).
  • standard math Rectified-flow interpolation and the MeanFlow identity (8) hold in the video latent space.
    Borrowed from [44,17]; used to define inner flow and training objective (Eqs. 5-9).
  • ad hoc to paper The final H DiT layers of the pretrained teacher can serve as a conditional inner-flow head while early-layer features remain fixed.
    Core architectural premise of TMD (Section 3); if false, the two-stage training cannot preserve the teacher distribution.
  • ad hoc to paper A single finite-difference step delta=0.005 approximates the total derivative d/ds u_theta well enough for training.
    Needed to compute Eq. (10) without JVP; accuracy is only validated empirically.
  • domain assumption The 500k synthetic videos generated by Wan2.1 14B are a representative training distribution for distilling both 1.3B and 14B students.
    Data table in Section 4.1 / Table 7; if synthetic distribution is biased, distillation inherits the bias.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transition Matching Distillation for Fast Video Generation." pith.science (2026). https://pith.science/paper/RXETR5JA

@misc{pith2026260109881,
  author       = {Pith},
  title        = {Pith review of: Transition Matching Distillation for Fast Video Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RXETR5JA}},
  note         = {Machine review of arXiv:2601.09881}
}
read the original abstract

Large video diffusion and flow models have achieved remarkable success in high-quality video generation, but their use in real-time interactive applications remains limited due to their inefficient multi-step sampling process. In this work, we present Transition Matching Distillation (TMD), a novel framework for distilling video diffusion models into efficient few-step generators. The central idea of TMD is to match the multi-step denoising trajectory of a diffusion model with a few-step probability transition process, where each transition is modeled as a lightweight conditional flow. To enable efficient distillation, we decompose the original diffusion backbone into two components: (1) a main backbone, comprising the majority of early layers, that extracts semantic representations at each outer transition step; and (2) a flow head, consisting of the last few layers, that leverages these representations to perform multiple inner flow updates. Given a pretrained video flow model, we first introduce a flow head to the model, and adapt it into a conditional flow map. We then apply distribution matching distillation to the student model with flow head rollout in each transition step. Extensive experiments on distilling Wan2.1 1.3B and 14B text-to-video models demonstrate that TMD provides a flexible and strong trade-off between generation speed and visual quality. In particular, TMD outperforms existing distilled models under comparable inference costs in terms of visual fidelity and prompt adherence. Project page: https://research.nvidia.com/labs/genair/tmd

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. DUET: A Diversity-Quality Duet of Distillation Experts for Two-Step Video Generation

    cs.CV 2026-08 conditional novelty 6.0 of 10

    DUET relays between an sCM expert at the high-noise step and a DMD expert at the low-noise step, obtaining both diversity and quality in two-step video generation, and DUET+ adds RL-based adaptation.

  2. OSVE: One Step Video Editing with One Step Diffusion Models

    cs.CV 2026-07 conditional novelty 6.0 of 10

    OSVE performs text-guided video editing with one-step diffusion models by training a single-pass inversion encoder and unifying frame latents for cross-frame attention, achieving quality comparable to multi-step metho...

  3. Salt: Self-Consistent Distribution Matching with Cache-Aware Training for Fast Video Generation

    cs.CV 2026-04 unverdicted novelty 6.0 of 10

    Self-consistent distribution matching plus cache-aware mixed-step training improves 2–4 NFE video quality on Wan 2.1 and real-time autoregressive backbones without extra inference cost.

Reference graph

Works this paper leans on

70 extracted references · 35 linked inside Pith · cited by 3 Pith papers

  1. [1]

    Cos- mos world foundation model platform for physical ai

    Niket Agarwal, Arslan Ali, Maciej Bala, Yogesh Bal- aji, Erik Barker, Tiffany Cai, Prithvijit Chattopad- hyay, Yongxin Chen, Yin Cui, Yifan Ding, et al. Cos- mos world foundation model platform for physical ai. arXiv preprint arXiv:2501.03575, 2025

    arXiv 2025

  2. [2]

    Building normalizing flows with stochastic inter- polants

    Michael Samuel Albergo and Eric Vanden-Eijnden. Building normalizing flows with stochastic inter- polants. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  3. [3]

    MoAlign: Motion-centric representation align- ment for video diffusion models.arXiv preprint arXiv:2510.19022, 2025

    Aritra Bhowmik, Denis Korzhenkov, Cees GM Snoek, Amirhossein Habibian, and Mohsen Ghafoo- rian. MoAlign: Motion-centric representation align- ment for video diffusion models.arXiv preprint arXiv:2510.19022, 2025

    arXiv 2025

  4. [4]

    Flow map matching.arXiv preprint arXiv:2406.07507, 2024

    Nicholas M Boffi, Michael S Albergo, and Eric Vanden-Eijnden. Flow map matching.arXiv preprint arXiv:2406.07507, 2024

    arXiv 2024

  5. [5]

    Video generation models as world simulators, 2024

    Tim Brooks, Bill Peebles, Connor Holmes, Will DePue, Yufei Guo, Li Jing, David Schnurr, Joe Taylor, Troy Luhman, Eric Luhman, Clarence Ng, Ricky Wang, and Aditya Ramesh. Video generation models as world simulators, 2024. openai.com/research/video-generation-models-as- world-simulators

    2024

  6. [6]

    Diffusion forcing: Next-token prediction meets full- sequence diffusion.Advances in Neural Information Processing Systems, 37:24081–24125, 2024

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

    2024

  7. [7]

    Videocrafter2: Overcoming data limitations for high- quality video diffusion models, 2024

    Haoxin Chen, Yong Zhang, Xiaodong Cun, Meng- han Xia, Xintao Wang, Chao Weng, and Ying Shan. Videocrafter2: Overcoming data limitations for high- quality video diffusion models, 2024

    2024

  8. [8]

    pi-flow: Policy-based few-step generation via imitation distil- lation, 2025

    Hansheng Chen, Kai Zhang, Hao Tan, Leonidas Guibas, Gordon Wetzstein, and Sai Bi. pi-flow: Policy-based few-step generation via imitation distil- lation, 2025

    2025

Show all 70 references
  1. [9]

    Sana-sprint: One-step diffusion with continuous-time consistency distillation.arXiv preprint arXiv:2503.09641, 2025

    Junsong Chen, Shuchen Xue, Yuyang Zhao, Jincheng Yu, Sayak Paul, Junyu Chen, Han Cai, Song Han, and Enze Xie. Sana-sprint: One-step diffusion with continuous-time consistency distillation.arXiv preprint arXiv:2503.09641, 2025

    2025

  2. [10]

    Sana- video: Efficient video generation with block linear dif- fusion transformer.arXiv preprint arXiv:2509.24695, 2025

    Junsong Chen, Yuyang Zhao, Jincheng Yu, Ruihang Chu, Junyu Chen, Shuai Yang, Xianbang Wang, Yicheng Pan, Daquan Zhou, Huan Ling, et al. Sana- video: Efficient video generation with block linear dif- fusion transformer.arXiv preprint arXiv:2509.24695, 2025

    2025

  3. [11]

    Self-forcing++: Towards minute- scale high-quality video generation.arXiv preprint arXiv:2510.02283, 2025

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

    2025 arXiv

  4. [12]

    Flashattention: Fast and memory- efficient exact attention with io-awareness.Advances in neural information processing systems, 35:16344– 16359, 2022

    Tri Dao, Dan Fu, Stefano Ermon, Atri Rudra, and Christopher Ré. Flashattention: Fast and memory- efficient exact attention with io-awareness.Advances in neural information processing systems, 35:16344– 16359, 2022

    2022

  5. [13]

    DOLLAR: Few-step video gen- eration via distillation and latent reward optimiza- tion

    Zihan Ding, Chi Jin, Difan Liu, Haitian Zheng, Kr- ishna Kumar Singh, Qiang Zhang, Yan Kang, Zhe Lin, and Yuchen Liu. DOLLAR: Few-step video gen- eration via distillation and latent reward optimiza- tion. InProceedings of the IEEE/CVF International Conference on Computer Visio...

    2025

  6. [14]

    One step diffusion via shortcut models

    Kevin Frans, Danijar Hafner, Sergey Levine, and Pieter Abbeel. One step diffusion via shortcut models. arXiv preprint arXiv:2410.12557, 2024

    2024 arXiv

  7. [15]

    Seedance 1.0: Explor- ing the boundaries of video generation models.arXiv preprint arXiv:2506.09113, 2025

    Yu Gao, Haoyuan Guo, Tuyen Hoang, Weilin Huang, Lu Jiang, Fangyuan Kong, Huixia Li, Jiashi Li, Liang Li, Xiaojie Li, et al. Seedance 1.0: Explor- ing the boundaries of video generation models.arXiv preprint arXiv:2506.09113, 2025

    2025 arXiv

  8. [16]

    Consistency models made easy.arXiv preprint arXiv:2406.14548, 2024

    Zhengyang Geng, Ashwini Pokle, William Luo, Justin Lin, and J Zico Kolter. Consistency models made easy.arXiv preprint arXiv:2406.14548, 2024

    2024 arXiv

  9. [17]

    Mean flows for one-step generative modeling.arXiv preprint arXiv:2505.13447, 2025

    Zhengyang Geng, Mingyang Deng, Xingjian Bai, J Zico Kolter, and Kaiming He. Mean flows for one-step generative modeling.arXiv preprint arXiv:2505.13447, 2025

    2025 arXiv

  10. [18]

    Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Communications of the ACM, 63(11):139– 144, 2020

    2020

  11. [19]

    Denois- ing diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

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

    2020

  12. [20]

    GLASS flows: Transition sampling for align- ment of flow and diffusion models.arXiv preprint arXiv:2509.25170, 2025

    Peter Holderrieth, Uriel Singer, Tommi Jaakkola, Ricky TQ Chen, Yaron Lipman, and Brian Kar- rer. GLASS flows: Transition sampling for align- ment of flow and diffusion models.arXiv preprint arXiv:2509.25170, 2025

    2025

  13. [21]

    Self forcing: Bridging the train- test gap in autoregressive video diffusion.arXiv preprint arXiv:2506.08009, 2025

    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. 10 Transition Matching Distillation for Fast Video Generation

    2025 arXiv

  14. [22]

    VBench: Comprehensive benchmark suite for video generative models

    Ziqi Huang, Yinan He, Jiashuo Yu, Fan Zhang, Chenyang Si, Yuming Jiang, Yuanhan Zhang, Tianx- ing Wu, Qingyang Jin, Nattapol Chanpaisit, Yao- hui Wang, Xinyuan Chen, Limin Wang, Dahua Lin, Yu Qiao, and Ziwei Liu. VBench: Comprehensive benchmark suite for video generative model...

    2024

  15. [23]

    Deepspeed ulysses: System optimizations for enabling training of extreme long sequence transformer models.arXiv preprint arXiv:2309.14509, 2023

    Sam Ade Jacobs, Masahiro Tanaka, Chengming Zhang, Minjia Zhang, Shuaiwen Leon Song, Samyam Rajbhandari, and Yuxiong He. Deepspeed ulysses: System optimizations for enabling training of extreme long sequence transformer models.arXiv preprint arXiv:2309.14509, 2023

    2023 arXiv

  16. [24]

    Consistency trajectory models: Learning probabil- ity flow ode trajectory of diffusion.arXiv preprint arXiv:2310.02279, 2023

    Dongjun Kim, Chieh-Hsin Lai, Wei-Hsiang Liao, Naoki Murata, Yuhta Takida, Toshimitsu Uesaka, Yutong He, Yuki Mitsufuji, and Stefano Ermon. Consistency trajectory models: Learning probabil- ity flow ode trajectory of diffusion.arXiv preprint arXiv:2310.02279, 2023

    2023 arXiv

  17. [25]

    Hunyuanvideo: A sys- tematic framework for large video generative models

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

    2024 arXiv

  18. [26]

    Trun- cated consistency models, 2024

    Sangyun Lee, Yilun Xu, Tomas Geffner, Giulia Fanti, Karsten Kreis, Arash Vahdat, and Weili Nie. Trun- cated consistency models, 2024

    2024

  19. [27]

    Advancing end-to-end pixel space generative model- ing via self-supervised pre-training.arXiv preprint arXiv:2510.12586, 2025

    Jiachen Lei, Keli Liu, Julius Berner, Haiming Yu, Hongkai Zheng, Jiahong Wu, and Xiangxiang Chu. Advancing end-to-end pixel space generative model- ing via self-supervised pre-training.arXiv preprint arXiv:2510.12586, 2025

    2025

  20. [28]

    T2v-turbo-v2: Enhancing video model post-training through data, reward, and con- ditional guidance design

    Jiachen Li, Qian Long, Jian Zheng, Xiaofeng Gao, Robinson Piramuthu, Wenhu Chen, and William Yang Wang. T2v-turbo-v2: Enhancing video model post-training through data, reward, and con- ditional guidance design. InThe Thirteenth Interna- tional Conference on Learning Representa...

    2025

  21. [29]

    Diffusion adversarial post-training for one-step video generation.arXiv preprint arXiv:2501.08316, 2025

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

    2025

  22. [30]

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

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

    2025

  23. [31]

    Flow match- ing for generative modeling.arXiv preprint arXiv:2210.02747, 2022

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

    2022 arXiv

  24. [32]

    Flow matching for generative modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InThe Eleventh Interna- tional Conference on Learning Representations, 2023

    2023

  25. [33]

    Fasterdiffusionviatemporalattentiondecomposition

    Haozhe Liu, Wentian Zhang, Jinheng Xie, Francesco Faccio, Mengmeng Xu, Tao Xiang, Mike Zheng Shou, Juan-Manuel Perez-Rua, and Jürgen Schmidhuber. Fasterdiffusionviatemporalattentiondecomposition. arXiv preprint arXiv:2404.02747, 2024

    2024 arXiv

  26. [34]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, et al. Flow straight and fast: Learning to generate and transfer data with rectified flow. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  27. [35]

    Simplifying, stabiliz- ing and scaling continuous-time consistency models

    Cheng Lu and Yang Song. Simplifying, stabiliz- ing and scaling continuous-time consistency models. arXiv preprint arXiv:2410.11081, 2024

    2024 arXiv

  28. [36]

    Knowledge distil- lation in iterative generative models for improved sampling speed.arXiv preprint arXiv:2101.02388, 2021

    Eric Luhman and Troy Luhman. Knowledge distil- lation in iterative generative models for improved sampling speed.arXiv preprint arXiv:2101.02388, 2021

    2021 arXiv

  29. [37]

    One-step diffusion distilla- tion through score implicit matching.Advances in Neural Information Processing Systems, 37:115377– 115408, 2024

    Weijian Luo, Zemin Huang, Zhengyang Geng, J Zico Kolter, and Guo-jun Qi. One-step diffusion distilla- tion through score implicit matching.Advances in Neural Information Processing Systems, 37:115377– 115408, 2024

    2024

  30. [38]

    Learning-to-cache: Accelerating diffusion transformer via layer caching.Advances in Neural Information Processing Systems, 37:133282–133304, 2024

    Xinyin Ma, Gongfan Fang, Michael Bi Mi, and Xin- chao Wang. Learning-to-cache: Accelerating diffusion transformer via layer caching.Advances in Neural Information Processing Systems, 37:133282–133304, 2024

    2024

  31. [39]

    Align your flow: Scaling continuous-time flow map distillation.arXiv preprint arXiv:2506.14603, 2025

    Amirmojtaba Sabour, Sanja Fidler, and Karsten Kreis. Align your flow: Scaling continuous-time flow map distillation.arXiv preprint arXiv:2506.14603, 2025

    2025 arXiv

  32. [40]

    Progressive distil- lation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512, 2022

    Tim Salimans and Jonathan Ho. Progressive distil- lation for fast sampling of diffusion models.arXiv preprint arXiv:2202.00512, 2022

    2022 arXiv

  33. [41]

    Fast high-resolution image synthesis with latent adversar- ial diffusion distillation

    Axel Sauer, Frederic Boesel, Tim Dockhorn, Andreas Blattmann, Patrick Esser, and Robin Rombach. Fast high-resolution image synthesis with latent adversar- ial diffusion distillation. InSIGGRAPH Asia 2024 Conference Papers, pages 1–11, 2024

    2024

  34. [42]

    Adversarial diffusion distilla- tion

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distilla- tion. InEuropean Conference on Computer Vision, pages 87–103. Springer, 2024

    2024

  35. [43]

    Seaweed-7b: Cost-effective training of video generation foundation model.arXiv preprint arXiv:2504.08685, 2025

    Team Seawead, Ceyuan Yang, Zhijie Lin, Yang Zhao, Shanchuan Lin, Zhibei Ma, Haoyuan Guo, Hao Chen, Lu Qi, Sen Wang, et al. Seaweed-7b: Cost-effective training of video generation foundation model.arXiv preprint arXiv:2504.08685, 2025

    2025 arXiv

  36. [44]

    Transition matching: Scalable and flexible generative modeling.arXiv preprint arXiv:2506.23589, 2025

    Neta Shaul, Uriel Singer, Itai Gat, and Yaron Lipman. Transition matching: Scalable and flexible generative modeling.arXiv preprint arXiv:2506.23589, 2025. 11 Transition Matching Distillation for Fast Video Generation

    2025 arXiv

  37. [45]

    Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020

    Yang 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

    2011 arXiv

  38. [46]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. InInternational Conference on Machine Learning, pages 32211–32252. PMLR, 2023

    2023

  39. [47]

    Unified continuous generative models.arXiv preprint arXiv:2505.07447, 2025

    Peng Sun, Yi Jiang, and Tao Lin. Unified continuous generative models.arXiv preprint arXiv:2505.07447, 2025

    2025 arXiv

  40. [48]

    Magi-1: Autore- gressive video generation at scale.arXiv preprint arXiv:2505.13211, 2025

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

    2025 arXiv

  41. [49]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

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

    2025 arXiv

  42. [50]

    DDT: Decoupled diffusion transformer.arXiv preprint arXiv:2504.05741, 2025

    Shuai Wang, Zhi Tian, Weilin Huang, and Limin Wang. DDT: Decoupled diffusion transformer.arXiv preprint arXiv:2504.05741, 2025

    2025 arXiv

  43. [51]

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

    Wenhao Wang and Yi Yang. 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

  44. [52]

    Transition models: Rethinking the generative learn- ing objective.arXiv preprint arXiv:2509.04394, 2025

    Zidong Wang, Yiyuan Zhang, Xiaoyu Yue, Xiangyu Yue, Yangguang Li, Wanli Ouyang, and Lei Bai. Transition models: Rethinking the generative learn- ing objective.arXiv preprint arXiv:2509.04394, 2025

    2025 arXiv

  45. [53]

    Sparse videogen: Accelerat- ingvideodiffusiontransformerswithspatial-temporal sparsity.arXiv preprint arXiv:2502.01776, 2025

    Haocheng Xi, Shuo Yang, Yilong Zhao, Chenfeng Xu, Muyang Li, Xiuyu Li, Yujun Lin, Han Cai, Jintao Zhang, Dacheng Li, et al. Sparse videogen: Accelerat- ingvideodiffusiontransformerswithspatial-temporal sparsity.arXiv preprint arXiv:2502.01776, 2025

    2025 arXiv

  46. [54]

    One-step dif- fusion models with𝑓-divergence distribution match- ing.arXiv preprint arXiv:2502.15681, 2025

    Yilun Xu, Weili Nie, and Arash Vahdat. One-step dif- fusion models with𝑓-divergence distribution match- ing.arXiv preprint arXiv:2502.15681, 2025

    2025 arXiv

  47. [55]

    Qwen2.5 technical report.arXiv preprint arXiv:2412.15115, 2024

    An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, et al. Qwen2.5 technical report.arXiv preprint arXiv:2412.15115, 2024

    2024 arXiv

  48. [56]

    Sparse videogen2: Accelerate video generation with sparse attention via semantic-aware permutation.arXiv preprint arXiv:2505.18875, 2025

    Shuo Yang, Haocheng Xi, Yilong Zhao, Muyang Li, Jintao Zhang, Han Cai, Yujun Lin, Xiuyu Li, Chenfeng Xu, Kelly Peng, et al. Sparse videogen2: Accelerate video generation with sparse attention via semantic-aware permutation.arXiv preprint arXiv:2505.18875, 2025

    2025 arXiv

  49. [57]

    Cogvideox: Text-to-video diffusion models with an expert transformer

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  50. [58]

    Improved distribution matching distillation for fast image synthesis

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

    2024

  51. [59]

    One-step diffusion with distribu- tion matching distillation

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribu- tion matching distillation. InProceedings of the IEEE/CVF conference on computer vision and pat- tern recognition, pages 6613–6623, 2024

    2024

  52. [60]

    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. InProceed- ings of the Computer Vision and Pattern Recognition Conference, pages 22963–22974, 2025

    2025

  53. [61]

    Representation alignment for genera- tion: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940, 2024

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for genera- tion: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940, 2024

    2024 arXiv

  54. [62]

    Packing input frame context in next-frame prediction models for video generation.arXiv preprint arXiv:2504.12626, 2025

    Lvmin Zhang and Maneesh Agrawala. Packing input frame context in next-frame prediction models for video generation.arXiv preprint arXiv:2504.12626, 2025

    2025

  55. [63]

    VideoREPA: Learning physics for video gen- eration through relational alignment with foundation models.arXiv preprint arXiv:2505.23656, 2025

    Xiangdong Zhang, Jiaqi Liao, Shaofeng Zhang, Fan- qing Meng, Xiangpeng Wan, Junchi Yan, and Yu Cheng. VideoREPA: Learning physics for video gen- eration through relational alignment with foundation models.arXiv preprint arXiv:2505.23656, 2025

    2025 arXiv

  56. [64]

    Metaxas, Sergey Tulyakov, and Jian Ren

    Zhixing Zhang, Yanyu Li, Yushu Wu, yanwu xu, Anil Kag, Ivan Skorokhodov, Willi Menapace, Aliaksandr Siarohin, Junli Cao, Dimitris N. Metaxas, Sergey Tulyakov, and Jian Ren. SF-v: Single forward video generation model. InThe Thirty-eighth Annual Con- ference on Neural Informati...

    2024

  57. [65]

    Real-time video generation with pyramid at- tention broadcast.arXiv preprint arXiv:2408.12588, 2024

    Xuanlei Zhao, Xiaolong Jin, Kai Wang, and Yang You. Real-time video generation with pyramid at- tention broadcast.arXiv preprint arXiv:2408.12588, 2024

    2024 arXiv

  58. [66]

    Pytorch fsdp: experiences on scaling fully sharded data par- allel.arXiv preprint arXiv:2304.11277, 2023

    Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, et al. Pytorch fsdp: experiences on scaling fully sharded data par- allel.arXiv preprint arXiv:2304.11277, 2023. 12 Transition Matching Distillation...

    2023 arXiv

  59. [67]

    Diffusion transformers with representa- tion autoencoders.arXiv preprint arXiv:2510.11690, 2025

    Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with representa- tion autoencoders.arXiv preprint arXiv:2510.11690, 2025

    2025 arXiv

  60. [68]

    Fast sam- pling of diffusion models via operator learning

    Hongkai Zheng, Weili Nie, Arash Vahdat, Kamyar Azizzadenesheli, and Anima Anandkumar. Fast sam- pling of diffusion models via operator learning. In International conference on machine learning, pages 42390–42402. PMLR, 2023

    2023

  61. [69]

    Large scale diffusion distillation via score-regularized continuous- time consistency.arXiv preprint arXiv: 2510.08431, 2025

    Kaiwen Zheng, Yuji Wang, Qianli Ma, Huayu Chen, Jintao Zhang, Yogesh 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

    2025 arXiv

  62. [70]

    gated” (b) Fusion type: “concat

    Mingyuan Zhou, Huangjie Zheng, Zhendong Wang, Mingzhang Yin, and Hai Huang. Score identity dis- tillation: Exponentially fast distillation of pretrained diffusion models for one-step generation. InForty- first International Conference on Machine Learning, 2024. 13 Transition M...

    2024

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.