pith. machine review for the scientific record. sign in

arxiv: 2604.04018 · v1 · submitted 2026-04-05 · 💻 cs.CV

Recognition: unknown

1.x-Distill: Breaking the Diversity, Quality, and Efficiency Barrier in Distribution Matching Distillation

Haoyu Li , Tingyan Wen , Lin Qi , Zhe Wu , Yihuang Chen , Xing Zhou , Lifei Zhu , Xueqian Wang
show 1 more author
Kai Zhang
Authors on Pith no claims yet

Pith reviewed 2026-05-13 17:16 UTC · model grok-4.3

classification 💻 cs.CV
keywords distribution matching distillationfew-step generationdiffusion modelstext-to-imagefractional stepsclassifier-free guidancemodel distillation
0
0 comments X

The pith

1.x-Distill enables practical 1.x-step generation in diffusion models by breaking the integer-step limit of prior distillation methods.

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

The paper introduces 1.x-Distill as the first fractional-step framework for distribution matching distillation of diffusion models. It modifies the teacher's classifier-free guidance to reduce mode collapse and applies a two-stage process that first preserves diversity through distribution matching before refining details via adversarial training. A compensation module also allows integration of caching during distillation. On SD3-Medium and SD3.5-Large, the method delivers higher quality and diversity than earlier few-step approaches at 1.67 and 1.74 effective steps while providing substantial speedups. This matters for making high-quality text-to-image generation faster and more efficient in practice.

Core claim

1.x-Distill establishes that distribution matching distillation can operate effectively in the fractional-step regime by first analyzing and adjusting the overlooked role of teacher CFG to suppress collapse, then using Stagewise Focused Distillation to separate coarse structure learning from detail refinement, and adding a lightweight compensation module to support block-level caching, resulting in better quality and diversity than integer-step baselines at 1.67 and 1.74 effective NFEs on SD3 models.

What carries the argument

The fractional-step distillation framework consisting of modified teacher CFG, Stagewise Focused Distillation (two-stage coarse-to-fine strategy), and a compensation module for Distill-Cache co-Training.

If this is right

  • Surpasses prior few-step distillation methods in quality and diversity at 1.67 effective NFEs on SD3-Medium.
  • Surpasses prior methods at 1.74 effective NFEs on SD3.5-Large.
  • Delivers up to 33x speedup relative to the original 28x2 NFE sampling.
  • Maintains diversity while achieving inference-consistent refinement through the two-stage process.
  • Naturally supports block-level caching inside the distillation pipeline.

Where Pith is reading between the lines

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

  • The fractional-step idea could be tested on non-SD3 diffusion architectures to check if the same CFG adjustment prevents collapse.
  • Real-time generation pipelines on edge devices might benefit if the caching module reduces memory traffic at these low step counts.
  • Similar two-stage separation of diversity matching from adversarial refinement could apply to other generative distillation tasks like video or audio.

Load-bearing premise

The proposed CFG modification and two-stage strategy will work on models beyond the tested SD3 variants without introducing new collapse modes or requiring heavy per-model retuning.

What would settle it

Running the full 1.x-Distill pipeline on a different base model such as SDXL and measuring whether quality or diversity metrics at 1.7 effective NFEs fall below those of a standard 2-step baseline would settle the claim.

Figures

Figures reproduced from arXiv: 2604.04018 by Haoyu Li, Kai Zhang, Lifei Zhu, Lin Qi, Tingyan Wen, Xing Zhou, Xueqian Wang, Yihuang Chen, Zhe Wu.

Figure 1
Figure 1. Figure 1: Visual results. 1.x-Distill mitigates the mode collapse and quality degrada￾tion of vanilla DMD under extreme step reduction, delivering superior few-step results. diffusion model into a few-step generator. Among them, Distribution matching distillation (DMD) [49, 50] reduces the student’s sampling to a few steps by matching the output distribution of the teacher, and have demonstrated strong effectiveness… view at source ↗
Figure 2
Figure 2. Figure 2: An illustration of the effect of the teacher’s CFG in distillation. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of 1.x-Distill. Our guidance control (Sec. 3.2) and cache design (Sec. 3.4) are both constructed in the two-stage framework (Sec. 3.3). Stage I: Train the generator with DMD loss. Within DMD framework, we apply importance sampling on diffusion timestep t, and control the guidance according to sampled t when compute the real score. Stage II: Train the generator with pixel-space adversarial loss. Ou… view at source ↗
Figure 4
Figure 4. Figure 4: Importance sampling in Stage I. Left: Under teacher scheduler (shift=3.0), we split timesteps from 1.0 to 0.0 into four windows to probe their effects. Right: Uniform sampling treats all timesteps equally, while our importance sampling down￾weights less informative ones and concentrates training on the more reliable region. refinement. For the discriminator, we follow the architecture in [20]. A frozen Con… view at source ↗
Figure 5
Figure 5. Figure 5: Caching for a distilled 2-step student. (a) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Qualitative comparison on SD3-Medium. Since most open-source base￾lines only provide 4-step checkpoints, we first compare all methods at 4 steps for fair￾ness. Our SFD already produces clearer and more appealing images, while other meth￾ods often show generation failures, color shifts, blur, or degraded aesthetics (red boxes). When pushed to 2 steps, 1.x-Distill still clearly surpasses all baselines. Quant… view at source ↗
Figure 7
Figure 7. Figure 7: User study: Comparing images generated by 1.x-Distill with other models. 4.3 Ablation Studies We perform comprehensive ablation studies to validate the effectiveness of each component and identify optimal design choices. Unless otherwise noted, all stud￾ies are performed in SD3-Medium at 1024 × 1024 [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Effect of guidance control strategy on sample diversity. Effect of Guidance Control To validate our guidance control strategy, we compare the sampling results of 1.x-Distill with and without it under a unified guidance scale w = 7.0. As shown in [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Effect of control threshold \alpha . 1.0 0.8 0.6 0.4 0.2 0.0 t 0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 Sampling Probability Uniform prob (w/o Prob-function) Prob function-a Prob function-b Prob function-c Prob function-d [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative ablation of block￾level caching for our DCT. As shown in Tab. 5 and [PITH_FULL_IMAGE:figures/full_fig_p014_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Effect of caching different block ranges. [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Effect of different training objectives in DCT. Experiments are conducted on SD3-Medium with a cache block 3–8, using identical training configurations. We compare different training objectives for DCT, including adversarial loss only (Ladv), adversarial loss with both KD objectives (Ladv + Lfeat + Lout), and the KD objectives individually. As shown in [PITH_FULL_IMAGE:figures/full_fig_p027_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Visual comparison of diversity under the 4-NFE setting distilled from SD3-Medium. Compared with two distribution-matching baselines, our approach pro￾duces more diverse samples while maintaining generation quality and prompt align￾ment [PITH_FULL_IMAGE:figures/full_fig_p029_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Additional qualitative comparison on SD3-Medium. [PITH_FULL_IMAGE:figures/full_fig_p030_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Additional qualitative comparison on SD3.5-Large. [PITH_FULL_IMAGE:figures/full_fig_p031_16.png] view at source ↗
read the original abstract

Diffusion models produce high-quality text-to-image results, but their iterative denoising is computationally expensive.Distribution Matching Distillation (DMD) emerges as a promising path to few-step distillation, but suffers from diversity collapse and fidelity degradation when reduced to two steps or fewer. We present 1.x-Distill, the first fractional-step distillation framework that breaks the integer-step constraint of prior few-step methods and establishes 1.x-step generation as a practical regime for distilled diffusion models.Specifically, we first analyze the overlooked role of teacher CFG in DMD and introduce a simple yet effective modification to suppress mode collapse. Then, to improve performance under extreme steps, we introduce Stagewise Focused Distillation, a two-stage strategy that learns coarse structure through diversity-preserving distribution matching and refines details with inference-consistent adversarial distillation. Furthermore, we design a lightweight compensation module for Distill--Cache co-Training, which naturally incorporates block-level caching into our distillation pipeline.Experiments on SD3-Medium and SD3.5-Large show that 1.x-Distill surpasses prior few-step methods, achieving better quality and diversity at 1.67 and 1.74 effective NFEs, respectively, with up to 33x speedup over original 28x2 NFE sampling.

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 / 1 minor

Summary. The paper introduces 1.x-Distill, the first fractional-step distillation framework for diffusion models that breaks the integer-step constraint of prior few-step methods. It modifies teacher CFG to suppress mode collapse, proposes Stagewise Focused Distillation (diversity-preserving distribution matching followed by inference-consistent adversarial refinement), and adds a lightweight compensation module for Distill-Cache co-training. On SD3-Medium and SD3.5-Large, it reports superior quality and diversity at 1.67 and 1.74 effective NFEs with up to 33x speedup over 28x2 NFE sampling.

Significance. If the empirical results hold under scrutiny, the work would establish 1.x-step generation as a practical regime for distilled diffusion models, addressing diversity collapse and fidelity issues in DMD at extreme low steps and delivering substantial efficiency gains for text-to-image synthesis.

major comments (2)
  1. [Experiments] Experiments section: The reported performance at 1.67 and 1.74 effective NFEs lacks error bars, ablation studies on the CFG modification strength and stagewise transition point, and statistical tests, making it impossible to assess robustness of the central quality/diversity claims.
  2. [Abstract and Experiments] Abstract and Experiments: The claim that the method establishes a 'practical regime' rests on results from only SD3-Medium and SD3.5-Large; no cross-architecture validation (e.g., U-Net backbones) is provided, leaving the generalization of the CFG tweak and two-stage curriculum untested and load-bearing for the broader contribution.
minor comments (1)
  1. [Method] The term '1.x-step' and 'effective NFEs' should be formally defined with an equation in the method section to avoid ambiguity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on experimental robustness and generalization. We have revised the manuscript to incorporate additional statistical analysis and a discussion of broader applicability, strengthening the central claims without overstating the current evidence.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The reported performance at 1.67 and 1.74 effective NFEs lacks error bars, ablation studies on the CFG modification strength and stagewise transition point, and statistical tests, making it impossible to assess robustness of the central quality/diversity claims.

    Authors: We agree that the original experiments would benefit from greater statistical rigor. In the revised manuscript we report error bars computed over five independent random seeds for all main-table metrics (FID, CLIP score, and diversity measures). We further add a dedicated ablation subsection that sweeps the CFG-modification strength over the range [0.5, 2.0] and the stagewise transition point at 1.5, 2.0, and 2.5 effective NFEs. Statistical significance between 1.x-Distill and the strongest baselines is assessed with paired Wilcoxon signed-rank tests; p-values are reported in the new Appendix C. These additions directly address the robustness concern. revision: yes

  2. Referee: [Abstract and Experiments] Abstract and Experiments: The claim that the method establishes a 'practical regime' rests on results from only SD3-Medium and SD3.5-Large; no cross-architecture validation (e.g., U-Net backbones) is provided, leaving the generalization of the CFG tweak and two-stage curriculum untested and load-bearing for the broader contribution.

    Authors: We acknowledge that the primary empirical support is limited to the SD3 family. The core algorithmic components—the CFG-suppression rule, the two-stage curriculum, and the lightweight compensation module—are formulated at the level of the teacher-student distribution-matching objective and are therefore architecture-agnostic. To make this explicit we have added a short paragraph in the revised Discussion section that explains why each component transfers to U-Net-based models, together with a compact set of preliminary SDXL results (U-Net backbone) placed in Appendix D. These results exhibit qualitatively similar gains in diversity and fidelity at sub-2 NFE regimes. Full-scale cross-architecture benchmarking remains computationally intensive and is noted as future work; the current evidence therefore supports the practical-regime claim for transformer-based diffusion models while leaving wider validation open. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation chain; empirical claims stand independently

full rationale

The paper introduces a modified teacher CFG and a two-stage Stagewise Focused Distillation strategy, then reports empirical results on SD3-Medium and SD3.5-Large at 1.67/1.74 effective NFEs. No equations, predictions, or central claims reduce by construction to fitted inputs, self-definitions, or load-bearing self-citations. The derivation chain consists of novel algorithmic modifications validated externally via experiments rather than tautological renaming or self-referential fitting.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claims rest on standard assumptions of diffusion model training (Gaussian noise schedule, CFG guidance) plus several new empirical choices whose justification is not derivable from first principles.

free parameters (2)
  • stagewise transition point
    The boundary between coarse diversity-preserving matching and detail-refinement adversarial stage is chosen empirically.
  • CFG modification strength
    The exact scaling or suppression factor applied to teacher CFG is tuned to suppress mode collapse.
axioms (1)
  • domain assumption Distribution matching loss preserves diversity when applied at coarse scale
    Invoked to justify the first stage of Stagewise Focused Distillation.

pith-pipeline@v0.9.0 · 5553 in / 1349 out tokens · 38723 ms · 2026-05-13T17:16:00.922696+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 1 Pith paper

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

  1. Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

    cs.CV 2026-05 unverdicted novelty 8.0

    CDM migrates distribution matching distillation to continuous time via dynamic random-length schedules and active off-trajectory latent alignment, yielding competitive few-step image fidelity on SD3 and Longcat-Image.

Reference graph

Works this paper leans on

53 extracted references · 53 canonical work pages · cited by 1 Pith paper · 11 internal anchors

  1. [1]

    5-flash: Distribution-guided distillation of generative flows

    Bandyopadhyay, H., Entezari, R., Scott, J., Adithyan, R., Song, Y.Z., Jampani, V.: Sd3. 5-flash: Distribution-guided distillation of generative flows. arXiv preprint arXiv:2509.21318 (2025) 3

    work page arXiv 2025

  2. [2]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Chadebec, C., Tasar, O., Benaroche, E., Aubin, B.: Flash diffusion: Accelerating any conditional diffusion model for few steps image generation. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 39, pp. 15686–15695 (2025) 10, 23

    work page 2025

  3. [3]

    δ-dit: A training-free acceleration method tailored for diffusion transformers.arXiv preprint arXiv:2406.01125, 2024

    Chen, P., Shen, M., Ye, P., Cao, J., Tu, C., Bouganis, C.S., Zhao, Y., Chen, T.: δ-dit: A training-free acceleration method tailored for diffusion transformers. arXiv preprint arXiv:2406.01125 (2024) 3, 4

    work page arXiv 2024

  4. [4]

    Advances in neural information processing systems34, 8780–8794 (2021) 1

    Dhariwal, P., Nichol, A.: Diffusion models beat gans on image synthesis. Advances in neural information processing systems34, 8780–8794 (2021) 1

    work page 2021

  5. [5]

    In: Forty-first international conference on machine learning (2024) 1, 9

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., et al.: Scaling rectified flow transformers for high-resolution image synthesis. In: Forty-first international conference on machine learning (2024) 1, 9

    work page 2024

  6. [6]

    arXiv preprint arXiv:2506.00523 (2025) 3

    Ge, X., Zhang, X., Xu, T., Zhang, Y., Zhang, X., Wang, Y., Zhang, J.: Sense- flow: Scaling distribution matching for flow-based text-to-image distillation. arXiv preprint arXiv:2506.00523 (2025) 2, 3, 4, 5, 7

    work page arXiv 2025

  7. [7]

    Mean Flows for One-step Generative Modeling

    Geng,Z.,Deng,M.,Bai,X.,Kolter,J.Z.,He,K.:Meanflowsforone-stepgenerative modeling. arXiv preprint arXiv:2505.13447 (2025) 1, 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Advances in neural infor- mation processing systems27(2014) 4, 7

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial nets. Advances in neural infor- mation processing systems27(2014) 4, 7

    work page 2014

  9. [9]

    Hessel, J., Holtzman, A., Forbes, M., Bras, R.L., Choi, Y.: Clipscore: A reference- freeevaluationmetricforimagecaptioning.arXivpreprintarXiv:2104.08718(2021) 10, 11, 23

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  10. [10]

    Advances in neural information processing systems30(2017) 10, 11, 23 16 H

    Heusel,M.,Ramsauer,H.,Unterthiner,T.,Nessler,B.,Hochreiter,S.:Ganstrained by a two time-scale update rule converge to a local nash equilibrium. Advances in neural information processing systems30(2017) 10, 11, 23 16 H. Li et al

    work page 2017

  11. [11]

    Advances in neural information processing systems33, 6840–6851 (2020) 1

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. Advances in neural information processing systems33, 6840–6851 (2020) 1

    work page 2020

  12. [12]

    Classifier-Free Diffusion Guidance

    Ho, J., Salimans, T.: Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598 (2022) 2, 5

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  13. [13]

    ELLA: Equip Diffusion Models with LLM for Enhanced Semantic Alignment

    Hu, X., Wang, R., Fang, Y., Fu, B., Cheng, P., Yu, G.: Ella: Equip diffusion models with llm for enhanced semantic alignment. arXiv preprint arXiv:2403.05135 (2024) 11, 23

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  14. [14]

    Distribution matching distillation meets reinforcement learning.arXiv preprint arXiv:2511.13649, 2025

    Jiang, D., Liu, D., Wang, Z., Wu, Q., Li, L., Li, H., Jin, X., Liu, D., Li, Z., Zhang, B., et al.: Distribution matching distillation meets reinforcement learning. arXiv preprint arXiv:2511.13649 (2025) 2

    work page arXiv 2025

  15. [15]

    Advances in neural information processing systems36, 36652–36663 (2023) 10, 11, 23

    Kirstain,Y.,Polyak,A.,Singer,U.,Matiana,S.,Penna,J.,Levy,O.:Pick-a-pic:An open dataset of user preferences for text-to-image generation. Advances in neural information processing systems36, 36652–36663 (2023) 10, 11, 23

    work page 2023

  16. [16]

    Advances in Neural Information Processing Systems37, 122458– 122483 (2024) 6

    Kynkäänniemi, T., Aittala, M., Karras, T., Laine, S., Aila, T., Lehtinen, J.: Ap- plying guidance in a limited interval improves sample and distribution quality in diffusion models. Advances in Neural Information Processing Systems37, 122458– 122483 (2024) 6

    work page 2024

  17. [17]

    Advances in neural information processing systems37 (2024) 4

    Li, S., Hu, T., van de Weijer, J., Khan, F.S., Liu, T., Li, L., Yang, S., Wang, Y., Cheng, M.M., Yang, J.: Faster diffusion: Rethinking the role of the encoder for diffusion model inference. Advances in neural information processing systems37 (2024) 4

    work page 2024

  18. [18]

    Sdxl- lightning: Progressive adversarial diffusion distillation

    Lin, S., Wang, A., Yang, X.: Sdxl-lightning: Progressive adversarial diffusion dis- tillation. arXiv preprint arXiv:2402.13929 (2024) 3, 4

    work page arXiv 2024

  19. [19]

    In: European conference on computer vision

    Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., Zitnick, C.L.: Microsoft coco: Common objects in context. In: European conference on computer vision. pp. 740–755. Springer (2014) 11

    work page 2014

  20. [20]

    ACM Transactions on Graphics (TOG)44(6), 1–21 (2025) 8, 21

    Lin, X., Yu, F., Hu, J., You, Z., Shi, W., Ren, J.S., Gu, J., Dong, C.: Harnessing diffusion-yielded score priors for image restoration. ACM Transactions on Graphics (TOG)44(6), 1–21 (2025) 8, 21

    work page 2025

  21. [21]

    Fastcache: Fast caching for diffusion transformer through learnable linear approximation.arXiv preprint arXiv:2505.20353, 2025

    Liu, D., Yu, Y., Zhang, J., Li, Y., Lengerich, B., Wu, Y.N.: Fastcache: Fast caching for diffusion transformer through learnable linear approximation. arXiv preprint arXiv:2505.20353 (2025) 4

    work page arXiv 2025

  22. [22]

    arXiv preprint arXiv:2511.22677 (2025) 4, 5

    Liu, D., Gao, P., Liu, D., Du, R., Li, Z., Wu, Q., Jin, X., Cao, S., Zhang, S., Li, H., et al.: Decoupled dmd: Cfg augmentation as the spear, distribution matching as the shield. arXiv preprint arXiv:2511.22677 (2025) 4, 5

    work page arXiv 2025

  23. [23]

    In: Proceedings of the Computer Vision and Pattern Recognition Conference

    Liu, F., Zhang, S., Wang, X., Wei, Y., Qiu, H., Zhao, Y., Zhang, Y., Ye, Q., Wan, F.: Timestep embedding tells: It’s time to cache for video diffusion model. In: Proceedings of the Computer Vision and Pattern Recognition Conference. pp. 7353–7363 (June 2025) 3, 4

    work page 2025

  24. [24]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Liu, J., Zou, C., Lyu, Y., Chen, J., Zhang, L.: From reusing to forecasting: Ac- celerating diffusion models with taylorseers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 15853–15863 (October 2025) 3, 4

    work page 2025

  25. [25]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Liu, Z., Mao, H., Wu, C.Y., Feichtenhofer, C., Darrell, T., Xie, S.: A convnet for the 2020s. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 11976–11986 (2022) 8

    work page 2022

  26. [26]

    arXiv preprint arXiv:2507.18569 (2025) 2, 4, 11 1.x-Distill 17

    Lu, Y., Ren, Y., Xia, X., Lin, S., Wang, X., Xiao, X., Ma, A.J., Xie, X., Lai, J.H.: Adversarial distribution matching for diffusion distillation towards efficient image and video synthesis. arXiv preprint arXiv:2507.18569 (2025) 2, 4, 11 1.x-Distill 17

    work page arXiv 2025

  27. [27]

    Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference

    Luo, S., Tan, Y., Huang, L., Li, J., Zhao, H.: Latent consistency mod- els: Synthesizing high-resolution images with few-step inference. arXiv preprint arXiv:2310.04378 (2023) 2, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  28. [28]

    Advances in Neural Information Processing Systems36, 76525–76546 (2023) 4

    Luo, W., Hu, T., Zhang, S., Sun, J., Li, Z., Zhang, Z.: Diff-instruct: A universal approach for transferring knowledge from pre-trained diffusion models. Advances in Neural Information Processing Systems36, 76525–76546 (2023) 4

    work page 2023

  29. [29]

    arXiv preprint arXiv:2503.06674 (2025) 2, 4, 5, 10, 23

    Luo, Y., Hu, T., Sun, J., Cai, Y., Tang, J.: Learning few-step diffusion models by trajectory distribution matching. arXiv preprint arXiv:2503.06674 (2025) 2, 4, 5, 10, 23

    work page arXiv 2025

  30. [30]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Ma, X., Fang, G., Wang, X.: Deepcache: Accelerating diffusion models for free. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 15762–15772 (2024) 4

    work page 2024

  31. [31]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Peebles, W., Xie, S.: Scalable diffusion models with transformers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 4195–4205 (2023) 4

    work page 2023

  32. [32]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952 (2023) 1

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  33. [33]

    arXiv preprint arXiv:2404.13686 (2024) 3

    Ren, Y., Xia, X., Lu, Y., Zhang, J., Wu, J., Xie, P., Wang, X., Xiao, X.: Hyper- sd: Trajectory segmented consistency model for efficient image synthesis. arXiv preprint arXiv:2404.13686 (2024) 1, 10, 23

    work page arXiv 2024

  34. [34]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 10684–10695 (2022) 1

    work page 2022

  35. [35]

    In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18

    Ronneberger, O., Fischer, P., Brox, T.: U-net: Convolutional networks for biomed- ical image segmentation. In: Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, Oc- tober 5-9, 2015, proceedings, part III 18. pp. 234–241. Springer (2015) 4

    work page 2015

  36. [36]

    Progressive Distillation for Fast Sampling of Diffusion Models

    Salimans, T., Ho, J.: Progressive distillation for fast sampling of diffusion models. arXiv preprint arXiv:2202.00512 (2022) 1, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  37. [37]

    In: SIGGRAPH Asia 2024 Conference Papers

    Sauer, A., Boesel, F., Dockhorn, T., Blattmann, A., Esser, P., Rombach, R.: Fast high-resolution image synthesis with latent adversarial diffusion distillation. In: SIGGRAPH Asia 2024 Conference Papers. pp. 1–11 (2024) 1, 3, 4, 10, 23

    work page 2024

  38. [38]

    In: European Conference on Computer Vision

    Sauer, A., Lorenz, D., Blattmann, A., Rombach, R.: Adversarial diffusion distilla- tion. In: European Conference on Computer Vision. pp. 87–103. Springer (2024) 1, 4

    work page 2024

  39. [39]

    Advances in neural information processing systems35, 25278–25294 (2022) 10, 11, 23

    Schuhmann, C., Beaumont, R., Vencu, R., Gordon, C., Wightman, R., Cherti, M., Coombes, T., Katta, A., Mullis, C., Wortsman, M., et al.: Laion-5b: An open large- scale dataset for training next generation image-text models. Advances in neural information processing systems35, 25278–25294 (2022) 10, 11, 23

    work page 2022

  40. [40]

    Fora: Fast-forward caching in diffusion transformer acceleration.arXiv preprint arXiv:2407.01425, 2024

    Selvaraju, P., Ding, T., Chen, T., Zharkov, I., Liang, L.: Fora: Fast-forward caching in diffusion transformer acceleration. arXiv preprint arXiv:2407.01425 (2024) 4

    work page arXiv 2024

  41. [41]

    In: Interna- tional Conference on Machine Learning

    Song, Y., Dhariwal, P., Chen, M., Sutskever, I.: Consistency models. In: Interna- tional Conference on Machine Learning. pp. 32211–32252. PMLR (2023) 1, 3

    work page 2023

  42. [42]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Song, Y., Sohl-Dickstein, J., Kingma, D.P., Kumar, A., Ermon, S., Poole, B.: Score- based generative modeling through stochastic differential equations. arXiv preprint arXiv:2011.13456 (2020) 5

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  43. [43]

    Advances in neural information processing systems36, 49659–49678 (2023) 10 18 H

    Sun, K., Pan, J., Ge, Y., Li, H., Duan, H., Wu, X., Zhang, R., Zhou, A., Qin, Z., Wang, Y., et al.: Journeydb: A benchmark for generative image understanding. Advances in neural information processing systems36, 49659–49678 (2023) 10 18 H. Li et al

    work page 2023

  44. [44]

    Advances in neural information processing systems37, 83951–84009 (2024) 10, 23

    Wang, F.Y., Huang, Z., Bergman, A., Shen, D., Gao, P., Lingelbach, M., Sun, K., Bian, W., Song, G., Liu, Y., et al.: Phased consistency models. Advances in neural information processing systems37, 83951–84009 (2024) 10, 23

    work page 2024

  45. [45]

    Advances in neural information processing systems36, 8406–8441 (2023) 4

    Wang, Z., Lu, C., Wang, Y., Bao, F., Li, C., Su, H., Zhu, J.: Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation. Advances in neural information processing systems36, 8406–8441 (2023) 4

    work page 2023

  46. [46]

    Qwen-Image Technical Report

    Wu, C., Li, J., Zhou, J., Lin, J., Gao, K., Yan, K., Yin, S.m., Bai, S., Xu, X., Chen, Y., et al.: Qwen-image technical report. arXiv preprint arXiv:2508.02324 (2025) 15

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  47. [47]

    Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesis

    Wu, X., Hao, Y., Sun, K., Chen, Y., Zhu, F., Zhao, R., Li, H.: Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis. arXiv preprint arXiv:2306.09341 (2023) 10, 11, 23

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  48. [48]

    Advances in Neural Information Processing Systems36, 15903–15935 (2023) 10, 11, 23

    Xu, J., Liu, X., Wu, Y., Tong, Y., Li, Q., Ding, M., Tang, J., Dong, Y.: Imagere- ward: Learning and evaluating human preferences for text-to-image generation. Advances in Neural Information Processing Systems36, 15903–15935 (2023) 10, 11, 23

    work page 2023

  49. [49]

    Advances in neural information processing systems37, 47455–47487 (2024) 1, 2, 4, 5, 7, 10

    Yin, T., Gharbi, M., Park, T., Zhang, R., Shechtman, E., Durand, F., Freeman, B.: Improved distribution matching distillation for fast image synthesis. Advances in neural information processing systems37, 47455–47487 (2024) 1, 2, 4, 5, 7, 10

    work page 2024

  50. [50]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Yin, T., Gharbi, M., Zhang, R., Shechtman, E., Durand, F., Freeman, W.T., Park, T.: One-step diffusion with distribution matching distillation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 6613–6623 (2024) 1, 2, 4, 11

    work page 2024

  51. [51]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 586–595 (2018) 11

    work page 2018

  52. [52]

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

    Zheng, K., Wang, Y., Ma, Q., Chen, H., Zhang, J., Balaji, Y., Chen, J., Liu, M.Y., Zhu, J., Zhang, Q.: Large scale diffusion distillation via score-regularized continuous-time consistency. arXiv preprint arXiv:2510.08431 (2025) 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  53. [53]

    arXiv preprint arXiv:2507.02860 (2025) 4 1.x-Distill 19 Appendix Table of Contents 1 Introduction

    Zhou, X., Liang, D., Chen, K., Feng, T., Chen, X., Lin, H., Ding, Y., Tan, F., Zhao, H., Bai, X.: Less is enough: Training-free video diffusion acceleration via runtime-adaptive caching. arXiv preprint arXiv:2507.02860 (2025) 4 1.x-Distill 19 Appendix Table of Contents 1 Introduction................................................... 1 2 Related Work .......

    work page arXiv 2025