Pith. sign in

REVIEW 3 major objections 5 minor 6 cited by

Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read EasyCache claims that reusing a cached transformation vector during locally stable denoising phases accelerates DiT video generation 2.1x–3.3x with no training and no offline profiling.

desk verdict A promising training-free caching method with broad empirical results, but the stated stability criterion does not actually control the reuse error as claimed. read the letter →

arxiv 2507.02860 v1 pith:WZIS3STU submitted 2025-07-03 cs.CV

Xin Zhou , Dingkang Liang , Kaijin Chen , Tianrui Feng , Xiwu Chen , Hongkai Lin , Yikang Ding , Feiyang Tan
show 2 more authors
Hengshuang Zhao Xiang Bai
This is my paper · ORCID
classification cs.CV
keywords videogenerationDiffusionTransformerinferenceaccelerationfeaturecachingtraining-freeruntime-adaptivedenoisingstabilitylatent
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

EasyCache claims that through most of the denoising trajectory of a Diffusion Transformer video model, the per-step transformation — the vector difference between the model's output and its input latent — is locally almost constant, so previously computed transformations can be reused instead of running the full network. The paper proposes a runtime-adaptive criterion that accumulates an estimated relative output-change rate and reuses the cached transformation only while the accumulated deviation stays below a threshold $\tau$. On OpenSora, Wan2.1, and HunyuanVideo this gives 2.12× to 2.54× speedups with no retraining and no offline profiling, rising to 3.33× when combined with an efficient-attention accelerator. If the claim is right, training-free caching becomes a plug-in technique that adapts per sample and per model, removing the dataset-specific tuning that limits prior caching schemes.

What carries the argument

The load-bearing object is the transformation vector $\Delta_t := v_t - x_t$ together with the accumulated stability indicator $E_t = \sum_{n=i+1}^{t} \varepsilon_n$, where each local indicator $\varepsilon_n$ approximates the output-change rate as $k_i\|x_n - x_{n-1}\| / \|v_{n-1}\|$ using the last fully computed rate $k_i$. When $E_t < \tau$, Eq. 6 reuses the cached $\Delta_i$; when the threshold is crossed, the full model runs, the cache is refreshed, and $E_t$ resets to zero. The warm-up count $R$ forces full computation during the early unstable phase, and the final step is always computed.

What would settle it

Record the actual per-step distances $\|\Delta_t - \Delta_i\|$ and the indicator $E_t$ across many prompts in a video DiT: if long stretches of later denoising show $E_t < \tau$ while $\|\Delta_t - \Delta_i\|$ grows enough to produce visible frame-to-frame artifacts or a VBench drop beyond 1%, the stability assumption and its proxy are falsified.

Watch

Extended reading notes

Core claim

The central discovery is that the relative transformation rate $k_t = \|v_t - v_{t-1}\| / \|x_t - x_{t-1}\|$ stabilizes quickly after a brief warm-up and stays nearly constant over most of the denoising process, so the transformation vector $\Delta_t := v_t - x_t$ can be approximated by a recently computed $\Delta_i$. EasyCache operationalizes this with the update rule $v_t = x_t + \Delta_i$ whenever the accumulated stability indicator $E_t$ stays below the threshold $\tau$, and a full forward pass otherwise, with the final step always fully computed. On HunyuanVideo this preserves PSNR 32.66 dB versus 23.85 dB for the previous dynamic-caching baseline at a higher speedup, and the same rule transfers to text-to-image generation.

Load-bearing premise

The load-bearing premise is that after a short warm-up the model's transformation vector stays nearly constant during denoising, and that the accumulated error estimate built from the last computed rate reliably detects when that constancy fails.

Editorial extensions

If this is right

  • On Open-Sora 1.2, Wan2.1-1.3B, and HunyuanVideo, EasyCache alone achieves 2.12×, 2.54×, and 2.21× speedups while improving PSNR, SSIM, and LPIPS over step-reduction, static-cache, PAB, and TeaCache at comparable settings.
  • On HunyuanVideo at 1280×720, EasyCache combined with SVG efficient attention reaches 3.33× total speedup with about a 1.1% PSNR drop, whereas the TeaCache+SVG combination loses 17.8% PSNR.
  • Because the criterion needs no offline profiling or dataset prior, a single $\tau, R$ setting transfers across video models and to text-to-image FLUX, where EasyCache reaches 4.64× speedup with FID 23.2 versus 25.8 for the full model.
  • VBench scores stay within 1% of the unaccelerated model on the tested prompts, indicating the quality loss is largely imperceptible to human evaluators.

Reading between the lines

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

  • A natural extension is to make $\tau$ schedule-dependent, allowing more aggressive reuse in the final refinement steps where visual differences are subtle; the paper only reports a single fixed threshold per model.
  • The stability assumption is qualitative, so a quantitative test would measure the distribution of $\|\Delta_t - \Delta_i\|$ across many prompts and check whether $E_t$ is a calibrated predictor of that distance.
  • Because EasyCache targets the DiT block's input/output transformation, it should compose with token-pruning or quantization accelerators, potentially closing part of the gap to real-time generation that the paper names as its main limitation.
  • The same constant-transformation-rate idea could be tested on other flow-matching or score-based samplers, including image or audio DiTs, where the cheap $E_t$ proxy is computable from the same quantities.
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

3 major / 5 minor

Summary. The paper proposes EasyCache, a training-free framework for accelerating DiT-based video diffusion models by reusing a cached transformation vector Δ_i = v_i − x_i at steps where a cumulative indicator E_t stays below a threshold τ. The indicator accumulates estimates of relative output change using a constant transformation rate k_i. Full model evaluations are triggered when E_t ≥ τ, during warm-up steps, and at the final step. Experiments on Open-Sora 1.2, Wan2.1-1.3B, HunyuanVideo, and FLUX.1-dev report speedups of 2.1–3.3× (up to 4.64× for text-to-image) with improved PSNR/SSIM/LPIPS over TeaCache and PAB, together with compatibility with the SVG attention acceleration. Ablations cover the threshold τ, warm-up steps R, the reuse criterion, and the update strategy for k.

Significance. If the underlying mechanism were established, this would be a practically valuable contribution: training-free, only two hyperparameters, model-agnostic across video and image DiTs, and accompanied by code. The empirical gains over strong baselines (e.g., PSNR 32.66 vs. 23.85 for TeaCache on HunyuanVideo at higher speedup) are substantial, and the ablations are informative. However, the paper's central theoretical claim—that the accumulated output-change indicator controls the reuse error of the cached transformation vector—is not supported by the submitted derivation, and the evaluation would benefit from matched-speedup comparisons. These issues are addressable in a revision but currently limit the scientific contribution.

major comments (3)
  1. [Sec. 3.3, Eqs. (4)-(6) and Sec. 3.4, Eq. (7)] The adaptive caching criterion does not control the actual reuse error ||Δ_t − Δ_i||. Under the Euler update in Eq. (7), Δ_t − Δ_{t−1} = (v_t − v_{t−1}) − v_{t−1} Δs_t; even when the output change v_t − v_{t−1} is negligible, the term v_{t−1} Δs_t causes Δ to drift at a rate of order Δs_t per step. The indicator E_t accumulates only an estimate of the relative output change, so it can remain near zero while the cached Δ drifts substantially (e.g., a constant-velocity field gives zero output change and unbounded Δ drift). The paper does not prove a bound on ||Δ_t − Δ_i|| in terms of E_t, and the claim in Sec. 3.3 that 'this stability, in turn, implies that the transformation vector Δ_t also remains nearly constant' is therefore not justified. Please either derive such a bound, modify the criterion to directly account for the step-size contribution (e.g., by accumulating Δs_t in E_t), or provide empirical evidence that E_t strongly correlates with the actual Δ drift across models and prompts.
  2. [Sec. 3.3, Eq. (4)] Under the approximation in Eq. (4) and the update in Eq. (7), ε_t ≈ k_i ||x_t − x_{t−1}|| / ||v_{t−1}|| = k_i Δs_t, since ||x_t − x_{t−1}|| = Δs_t ||v_{t−1}||. Consequently, the accumulated indicator E_t is approximately k_i times the elapsed continuous time since the last full computation, independent of the actual output dynamics. This makes the 'runtime-adaptive' criterion effectively a fixed time budget with a model-specific rate, which is a different mechanism from the local-stability detection described in the paper. The authors should acknowledge this interpretation, or change the indicator so that it does not cancel the velocity norm, otherwise the adaptive claim is overstated.
  3. [Table 1, Sec. 4.2] The headline PSNR gains compare methods at different speedups (e.g., HunyuanVideo: TeaCache at 1.67× with PSNR 23.85 vs. EasyCache at 2.21× with PSNR 32.66). While a Pareto improvement is strong, the comparison would be more conclusive with matched-speedup runs or full trade-off curves for all three video models. The τ-ablation in Table 3 provides such a curve only for Wan2.1-1.3B and with a single seed per prompt; extending this analysis and reporting variance across seeds would substantially strengthen the quality claim.
minor comments (5)
  1. [Abstract] There is a grammatical error: 'a efficient' should be 'an efficient'.
  2. [Fig. 2] The figure does not state which model, sampling configuration, and prompt are used for the analysis; the caption should specify these details and define the averaging operation used for the L1 norm.
  3. [Sec. 4.1] The paper states that five samples per VBench prompt are generated, but it does not report the number of prompts or standard deviations for PSNR, SSIM, and LPIPS. Please include these, as the ablations in Sec. 4.4 use a single sample per prompt and are therefore not statistically robust.
  4. [Eq. (4)] The definition of ε_t should make the percentage normalization explicit; the threshold values (e.g., τ = 5%) are given in percent, but the equation as written is a dimensionless ratio.
  5. [Sec. 4.2] The text states that the VBench drop is 'negligible' and 'below 1%', but the table shows Open-Sora 79.40 to 78.74 (−0.66) and Wan2.1 81.30 to 80.49 (−0.81); please present the percentage changes explicitly and discuss whether these are within the benchmark's run-to-run variability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the caching rule is a measured heuristic checked against external baselines, and the Δ-drift critique concerns correctness rather than definitional reduction.

full rationale

EasyCache's runtime decision rule (Eq. 6) reuses the cached transformation vector Δ_i whenever the accumulated indicator E_t remains below τ. The indicator is defined independently of the target quantity: E_t accumulates estimates of ||v_t − v_{t−1}||/||v_{t−1}|| built from k_i and observed input differences (Eq. 4-5), while the cached object is Δ_t = v_t − x_t (Eq. 3). Reuse is therefore a heuristic proxy, not a quantity fitted so that E_t and ||Δ_t − Δ_i|| coincide by construction. The central premise—transformation-rate stability—is presented as an empirical observation (Fig. 2b) and is not imported from a self-citation; references to TeaCache, PAB, and SVG are external baselines used for comparison. Hyperparameters (τ, R) are tuned per model in Tables 3-4, but this is ordinary threshold selection followed by independent measurement on held prompts, not fitting a parameter to the reported PSNR and then calling it a prediction. The paper's own Sec. 3.4 labels the stability argument an 'Intuitive Perspective' and gives no formal bound; the observation that Δ can drift with v_{t−1}Δs_t under the Euler update is a critique of the heuristic's correctness, not evidence that the claimed derivation reduces to its inputs by definition. The appended Limitation section addresses the real-time gap rather than this heuristic gap; that omission is a correctness/completeness concern, outside the circularity definition used here.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No invented physical entities. The method introduces a new caching criterion with two free hyperparameters per model and relies on empirical stability assumptions about diffusion transformer dynamics.

free parameters (2)
  • cache tolerance τ = 10%, 5%, 2.5%
    Hand-tuned per model in Sec.4.1; controls aggressiveness of reuse, directly trades speed for quality (Tab.3).
  • warm-up steps R = 5, 10, 5
    Hand-tuned per model; fully computed initial steps to avoid unstable transformation rates (Tab.4).
assumptions (4)
  • standard math Flow matching Euler update x_t = x_{t-1} + v_{t-1}Δs_t
    Used in Sec.3.4 to interpret stability of k_t.
  • domain assumption Transformation rate k_t is stable after R warm-up steps
    Empirical observation from Fig.2, used to justify constant-k approximation in Eq.4.
  • domain assumption Small output change implies the transformation vector Δ_t is nearly constant
    Core hypothesis of Sec.3.3; enables reuse in Eq.3.
  • domain assumption Accumulated estimated deviations E_t reliably indicate approximation error
    Reuse condition E_t < τ in Eq.6; assumed but not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching." pith.science (2026). https://pith.science/paper/WZIS3STU

@misc{pith2026250702860,
  author       = {Pith},
  title        = {Pith review of: Less is Enough: Training-Free Video Diffusion Acceleration via Runtime-Adaptive Caching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WZIS3STU}},
  note         = {Machine review of arXiv:2507.02860}
}
abstract

Video generation models have demonstrated remarkable performance, yet their broader adoption remains constrained by slow inference speeds and substantial computational costs, primarily due to the iterative nature of the denoising process. Addressing this bottleneck is essential for democratizing advanced video synthesis technologies and enabling their integration into real-world applications. This work proposes EasyCache, a training-free acceleration framework for video diffusion models. EasyCache introduces a lightweight, runtime-adaptive caching mechanism that dynamically reuses previously computed transformation vectors, avoiding redundant computations during inference. Unlike prior approaches, EasyCache requires no offline profiling, pre-computation, or extensive parameter tuning. We conduct comprehensive studies on various large-scale video generation models, including OpenSora, Wan2.1, and HunyuanVideo. Our method achieves leading acceleration performance, reducing inference time by up to 2.1-3.3$\times$ compared to the original baselines while maintaining high visual fidelity with a significant up to 36% PSNR improvement compared to the previous SOTA method. This improvement makes our EasyCache a efficient and highly accessible solution for high-quality video generation in both research and practical applications. The code is available at https://github.com/H-EmbodVis/EasyCache.

Figures

Figures reproduced from arXiv: 2507.02860 by the authors.

Figure 1
Figure 1. The comparison between (a) the default iterative de [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Analysis of feature dynamics. (a) The L1 norm of the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The overall of our method. For simplicity, we start from a computed step [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Qualitative comparison of EasyCache with baseline and prior acceleration methods [ [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: The compatibility with other acceleration techniques [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

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

  1. GVCCTurbo: Rate-Compute Quality Scheduling for Codebook Driven Generative Compression

    cs.CV 2026-08 conditional novelty 6.0 of 10

    GVCCTurbo reuses cached generator endpoints across codebook correction steps, cutting denoiser evaluations from 20 to 9 and decoding time by roughly 44% at similar LPIPS.

  2. WorldDynCache: Risk-Controlled Latent Dynamics Approximation for Diffusion World Model

    cs.LG 2026-08 conditional novelty 6.0 of 10

    A caching framework that approximates latent transitions with a condition-aware lifted surrogate and gates them by an online-calibrated, future-weighted risk, reporting 4.92x and 2.15x speedups on two diffusion world models.

  3. Sol-Attn: Accelerating Video Generation Inference via On-the-Fly Attention Sparsification

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Sol-Attn fuses threshold-based block selection and approximate correction into a single online-softmax pass, reporting 1.9–3.0× end-to-end speedups for video generation and editing without retraining.

  4. CachedSearch: Training-Free Cached Exploration for Test-Time Search in Video Diffusion

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Using cached drafts to select the winner and regenerating only that winner captures 94.7% of best-of-8 search gain at 63% of the cost.

  5. OmniCache: Multidimensional Hierarchical Feature Caching For Diffusion Models

    cs.CV 2026-07 conditional novelty 5.0 of 10

    Training-free hierarchical feature caching across token, frame, block, and layer axes cuts diffusion inference latency up to 35% while preserving quality better than averaging-based token merging.

  6. ACID: Adaptive Caching for vIDeo generation

    cs.CV 2026-07 conditional novelty 5.0 of 10

    Adaptive threshold switching on the drift signal's rate of change expands the quality-vs-speed Pareto frontier of TeaCache, EasyCache, and DiCache across HunyuanVideo, Wan 2.1, and CogVideoX.

Reference graph

Works this paper leans on

44 extracted references · 33 canonical work pages · cited by 6 Pith papers

  1. [1]

    Open-sora: Democratizing efficient video production for all,

    Z. Zheng, X. Peng, T. Yang, C. Shen, S. Li, H. Liu, Y. Zhou, T. Li, and Y. You, “Open-sora: Democratizing efficient video production for all,” arXiv preprint arXiv:2412.20404, 2024

    arXiv 2024

  2. [2]

    Latte: Latent diffusion transformer for video generation,

    X. Ma, Y. Wang, G. Jia, X. Chen, Z. Liu, Y.-F. Li, C. Chen, and Y. Qiao, “Latte: Latent diffusion transformer for video generation,” Trans. Mach. Learn. Research, 2024

    work page 2024

  3. [3]

    Video generation models as world simulators,

    T. Brooks et al., “Video generation models as world simulators,” OpenAI Blog, vol. 1, p. 8, 2024

    work page 2024

  4. [5]

    Scalable diffusion models with transform- ers,

    W. Peebles and S. Xie, “Scalable diffusion models with transform- ers,” in Proc. IEEE Int. Conf. Comput. Vis., 2023, pp. 4195–4205

    work page 2023

  5. [6]

    Hunyuanvideo: A systematic framework for large video generative models,

    W. Kong et al., “Hunyuanvideo: A systematic framework for large video generative models,” arXiv preprint arXiv:2412.03603, 2024

    arXiv 2024

  6. [7]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps,

    C. Lu, Y. Zhou, F. Bao, J. Chen, C. Li, and J. Zhu, “Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps,” in Proc. Neural Inf. Process. Syst. , vol. 35, 2022, pp. 5775–5787

    work page 2022

  7. [8]

    Reducing activation recomputa- tion in large transformer models,

    V . A. Korthikanti, J. Casper, S. Lym, L. McAfee, M. Andersch, M. Shoeybi, and B. Catanzaro, “Reducing activation recomputa- tion in large transformer models,” Proc. Mach. Learn. Syst. , vol. 5, pp. 341–353, 2023

    work page 2023

  8. [9]

    On distillation of guided diffusion models,

    C. Meng, R. Rombach, R. Gao, D. Kingma, S. Ermon, J. Ho, and T. Salimans, “On distillation of guided diffusion models,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2023, pp. 14 297–14 306

    work page 2023

Show all 44 references
  1. [10]

    Sageattention: Accurate 8-bit attention for plug-and-play inference acceleration,

    J. Zhang, J. Wei, P . Zhang, J. Zhu, and J. Chen, “Sageattention: Accurate 8-bit attention for plug-and-play inference acceleration,” in Proc. Int. Conf. Learn. Representations, 2025

    2025

  2. [11]

    Deepcache: Principled cache for mobile deep vision,

    M. Xu, M. Zhu, Y. Liu, F. X. Lin, and X. Liu, “Deepcache: Principled cache for mobile deep vision,” in Proc. Annual Int. conf. Mobile Comput. and Network., 2018, pp. 129–144

    2018

  3. [12]

    Real-time video genera- tion with pyramid attention broadcast,

    X. Zhao, X. Jin, K. Wang, and Y. You, “Real-time video genera- tion with pyramid attention broadcast,” in Proc. Int. Conf. Learn. Representations, 2025

    2025

  4. [13]

    δ-dit: A training-free acceleration method tailored for diffusion transformers,

    P . Chen, M. Shen, P . Ye, J. Cao, C. Tu, C.-S. Bouganis, Y. Zhao, and T. Chen, “ δ-dit: A training-free acceleration method tailored for diffusion transformers,” arXiv preprint arXiv:2406.01125, 2024

    2024 arXiv

  5. [14]

    Fora: Fast-forward caching in diffusion transformer acceleration,

    P . Selvaraju, T. Ding, T. Chen, I. Zharkov, and L. Liang, “Fora: Fast-forward caching in diffusion transformer acceleration,” arXiv preprint arXiv:2407.01425, 2024

    2024 arXiv

  6. [15]

    Timestep embedding tells: It’s time to cache for video diffusion model,

    F. Liu, S. Zhang, X. Wang, Y. Wei, H. Qiu, Y. Zhao, Y. Zhang, Q. Ye, and F. Wan, “Timestep embedding tells: It’s time to cache for video diffusion model,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2025, pp. 7353–7363

    2025

  7. [16]

    Sparse videogen: Accelerating video diffusion trans- formers with spatial-temporal sparsity,

    H. Xi et al., “Sparse videogen: Accelerating video diffusion trans- formers with spatial-temporal sparsity,” in Proc. Int. Conf. Mach. Learn., 2025

    2025

  8. [17]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P . Abbeel, “Denoising diffusion probabilistic models,” in Proc. Neural Inf. Process. Syst., 2020, pp. 6840–6851

    2020

  9. [18]

    Photorealistic text-to-image diffusion models with deep language understanding,

    C. Saharia et al. , “Photorealistic text-to-image diffusion models with deep language understanding,” in Proc. Neural Inf. Process. Syst., vol. 35, 2022, pp. 36 479–36 494

    2022

  10. [19]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P . Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2022, pp. 10 684– 10 695

    2022

  11. [20]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis,

    D. Podell, Z. English, K. Lacey, A. Blattmann, T. Dockhorn, J. M ¨uller, J. Penna, and R. Rombach, “Sdxl: Improving latent diffusion models for high-resolution image synthesis,” in Proc. Int. Conf. Learn. Representations, 2024

    2024

  12. [21]

    Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation,

    Z. Zhao et al. , “Hunyuan3d 2.0: Scaling diffusion models for high resolution textured 3d assets generation,” arXiv preprint arXiv:2501.12202, 2025

    2025 arXiv

  13. [22]

    Step1x-3d: Towards high-fidelity and controllable generation of textured 3d assets,

    W. Li et al. , “Step1x-3d: Towards high-fidelity and controllable generation of textured 3d assets,” arXiv preprint arXiv:2505.07747, 2025

    2025 arXiv

  14. [23]

    3dtopia-xl: Scaling high-quality 3d asset generation via primitive diffusion,

    Z. Chen et al., “3dtopia-xl: Scaling high-quality 3d asset generation via primitive diffusion,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2025, pp. 26 576–26 586

    2025

  15. [24]

    Mag- icvideo: Efficient video generation with latent diffusion models,

    D. Zhou, W. Wang, H. Yan, W. Lv, Y. Zhu, and J. Feng, “Mag- icvideo: Efficient video generation with latent diffusion models,” arXiv preprint arXiv:2211.11018, 2022

    2022 arXiv

  16. [25]

    Align your latents: High-resolution video synthesis with latent diffusion models,

    A. Blattmann, R. Rombach, H. Ling, T. Dockhorn, S. W. Kim, S. Fidler, and K. Kreis, “Align your latents: High-resolution video synthesis with latent diffusion models,” in Proc. IEEE Conf. Com- put. Vis. Pattern Recognit., 2023, pp. 22 563–22 575

    2023

  17. [26]

    Stable video diffusion: Scaling la- tent video diffusion models to large datasets,

    A. Blattmann et al. , “Stable video diffusion: Scaling la- tent video diffusion models to large datasets,” arXiv preprint arXiv:2311.15127, 2023

    2023 arXiv

  18. [27]

    Cogvideo: Large-scale pretraining for text-to-video generation via transform- ers,

    W. Hong, M. Ding, W. Zheng, X. Liu, and J. Tang, “Cogvideo: Large-scale pretraining for text-to-video generation via transform- ers,” in Proc. Int. Conf. Learn. Representations, 2023

    2023

  19. [28]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,” in Proc. Int. Conf. Learn. Representations, 2021

    2021

  20. [29]

    Elucidating the design space of diffusion-based generative models,

    T. Karras, M. Aittala, T. Aila, and S. Laine, “Elucidating the design space of diffusion-based generative models,” in Proc. Neural Inf. Process. Syst., vol. 35, 2022, pp. 26 565–26 577

    2022

  21. [30]

    Token merging for fast stable diffusion,

    D. Bolya and J. Hoffman, “Token merging for fast stable diffusion,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2023, pp. 4599– 4603

    2023

  22. [31]

    Attention-driven training-free efficiency enhancement of diffu- sion models,

    H. Wang, D. Liu, Y. Kang, Y. Li, Z. Lin, N. K. Jha, and Y. Liu, “Attention-driven training-free efficiency enhancement of diffu- sion models,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2024, pp. 16 080–16 089

    2024

  23. [32]

    Training-free and hardware-friendly acceleration for diffusion models via similarity- based token pruning,

    E. Zhang, J. Tang, X. Ning, and L. Zhang, “Training-free and hardware-friendly acceleration for diffusion models via similarity- based token pruning,” in Proc. AAAI Conf. Artif. Intell., 2025

    2025

  24. [33]

    Accelerating diffusion transformers with token-wise feature caching,

    C. Zou, X. Liu, T. Liu, S. Huang, and L. Zhang, “Accelerating diffusion transformers with token-wise feature caching,” in Proc. Int. Conf. Learn. Representations, 2025

    2025

  25. [34]

    Cache me if you can: Accelerating diffusion models through block caching,

    F. Wimbauer et al., “Cache me if you can: Accelerating diffusion models through block caching,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2024, pp. 6211–6220

    2024

  26. [35]

    Score-based generative modeling through stochastic differential equations,

    Y. Song, J. Sohl-Dickstein, D. P . Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differential equations,” in Proc. Int. Conf. Learn. Representations , 2021

    2021

  27. [36]

    Flow matching for generative modeling,

    Y. Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” in Proc. Int. Conf. Learn. Representations, 2023

    2023

  28. [37]

    Faster diffusion via temporal attention decomposition,

    H. Liu, W. Zhang, J. Xie, F. Faccio, M. Xu, T. Xiang, M. Z. Shou, J.- M. Perez-Rua, and J. Schmidhuber, “Faster diffusion via temporal attention decomposition,” Trans. Mach. Learn. Research, 2025

    2025

  29. [38]

    Wan: Open and advanced large-scale video generative models,

    A. Wang et al. , “Wan: Open and advanced large-scale video generative models,” arXiv preprint arXiv:2503.20314, 2025

    2025 arXiv

  30. [39]

    B. F. Labs, “Flux,” https://github.com/black-forest-labs/flux, 2024

    2024

  31. [40]

    Vbench: Comprehensive benchmark suite for video generative models,

    Z. Huang et al. , “Vbench: Comprehensive benchmark suite for video generative models,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2024, pp. 21 807–21 818

    2024

  32. [41]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P . Simoncelli, “Image quality assessment: from error visibility to structural similarity,” IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, 2004

    2004

  33. [42]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P . Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2018, pp. 586–595

    2018

  34. [43]

    Microsoft coco: Common objects in context,

    T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P . Perona, D. Ramanan, P . Doll´ar, and C. L. Zitnick, “Microsoft coco: Common objects in context,” in Proc. Eur. Conf. Comput. Vis., 2014, pp. 740–755

    2014

  35. [44]

    Clip- score: A reference-free evaluation metric for image captioning,

    J. Hessel, A. Holtzman, M. Forbes, R. Le Bras, and Y. Choi, “Clip- score: A reference-free evaluation metric for image captioning,” in Proc. Conf. Empirical Methods in Natural Language Process., 2021, pp. 7514–7528

    2021

  36. [45]

    T2v-compbench: A comprehensive benchmark for compositional text-to-video generation,

    K. Sun, K. Huang, X. Liu, Y. Wu, Z. Xu, Z. Li, and X. Liu, “T2v-compbench: A comprehensive benchmark for compositional text-to-video generation,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2025, pp. 8406–8416

    2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.