Pith. sign in

REVIEW 4 major objections 7 minor 4 cited by

Efficient Diffusion Models: A Survey

T0 review · 4 major / 7 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper's contribution is a three-part taxonomy—algorithm-level, system-level, and frameworks—that organizes the literature on efficient diffusion models and claims to be a reliable map for researchers and practitioners.

desk verdict Useful entry-point survey with a sensible taxonomy, but the 'systematic and comprehensive' claim is unsupported until a selection methodology is added and a few self-admitted gaps are fixed. read the letter →

arxiv 2502.06805 v3 pith:BCSS6662 submitted 2025-02-03 cs.LG cs.GR

Hui Shen , Jingxuan Zhang , Boning Xiong , Rui Hu , Shoufa Chen , Zhongwei Wan , Xin Wang , Yu Zhang
show 6 more authors
Zixuan Gong Guangyin Bao Chaofan Tao Yongfeng Huang Ye Yuan Mi Zhang
This is my paper · ORCID
classification cs.LGcs.GR
keywords efficientdiffusionmodelssurveytaxonomyalgorithm-leveloptimizationsystem-levelframeworksinferenceaccelerationmodelcompression
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 tries to establish that the scattered research on making diffusion models cheaper can be organized into a three-part taxonomy: algorithm-level techniques that change training, fine-tuning, sampling, or model size; system-level techniques that optimize hardware, parallelism, and caching; and specialized software frameworks that accelerate training or inference. A sympathetic reader would care because diffusion models are powerful but expensive, and the survey's value is giving practitioners a structured way to locate the right speedup technique instead of reading an ever-growing pile of papers. The paper also argues that this map is broader and more current than the one existing efficiency-focused survey, and it identifies open directions such as replacing classifier-free guidance and making attention efficient for video. The whole edifice rests on trusting the surveyed papers' reported speedups and quality metrics.

What carries the argument

The central object is the taxonomy itself: algorithm-level efficiency with subcategories for efficient training, fine-tuning, sampling, and compression; system-level efficiency with subcategories for hardware-software co-design, parallel computing, and caching; and frameworks with subcategories for training support and inference acceleration. The taxonomy carries the argument by assigning each surveyed technique a slot, which makes coverage, gaps, and connections visible and supports the survey's claim that the field is mature enough to map but still has open research directions.

What would settle it

Run a representative method from each taxonomy branch on identical hardware and compare measured speedups with the reported ones; alternatively, take a recent fixed set of diffusion-efficiency papers and count how many fail to fit exactly one taxonomy category.

Watch

Extended reading notes

Core claim

The central claim is that the many independent efforts to make diffusion models efficient form a coherent design space rather than a random collection. At the algorithm level, efficiency comes from latent-space training, alternative loss formulations such as rectified flow, adaptive priors, noise schedules, parameter-efficient fine-tuning, faster solvers, sampling schedules, distillation, quantization, and pruning. At the system level, it comes from hardware-software co-design, parallel inference and training, and caching of intermediate features. At the framework level, it comes from libraries that support both training and inference and from tools that optimize inference alone. The paper's contribution is to place hundreds of recent papers into this taxonomy and to point out where the field is heading.

Load-bearing premise

The map's reliability depends on the surveyed papers' reported speedups and quality numbers being accurate, and on each paper being assignable to exactly one taxonomy category.

Editorial extensions

If this is right

  • Practitioners facing slow sampling can look first to efficient solvers, sampling schedules, or distillation, since these require no system-level changes.
  • A model that is memory-bound can be quantized or pruned, and those compression techniques are largely orthogonal to sampling acceleration.
  • System-level techniques such as parallel inference and caching target deployment scale, so teams serving many users can combine them with algorithm-level speedups.
  • The identified future directions—hybrid diffusion and autoregressive models, guidance without classifier-free guidance, and sparse attention for video—mark where the next efficiency gains are expected.

Reading between the lines

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

  • The three categories are best read as a decision aid rather than a rigid partition: some methods, such as caching libraries, appear at both the system and framework levels.
  • If the reported speedups hold up, the practical bottleneck is choosing compatible combinations of techniques, not finding new tricks.
  • The future-work section suggests that the next wave of gains may come from changing the training objective rather than from better samplers or systems alone.
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

4 major / 7 minor

Summary. This manuscript is a survey of techniques for making diffusion models more efficient, organized into three advertised categories: algorithm-level methods (efficient training, fine-tuning, sampling, and compression), system-level methods (hardware-software co-design, parallel computing, and caching), and purpose-built frameworks. It provides background formulas for DDPM, score matching, SDE/ODE solvers, and flow matching, then surveys each of the three areas in turn, with a GitHub repository listing the covered papers. The abstract and introduction state that the review is 'systematic and comprehensive.' The paper makes no new algorithmic or empirical contributions; its value lies in the organization and description of existing work.

Significance. If the coverage claim were backed by a reproducible methodology, this survey would be a useful entry point for researchers and practitioners: the three-part taxonomy is reasonable, the description of the surveyed methods is mostly consistent with the cited literature, and the maintained GitHub repository is a practical supplement. The paper is also careful in some respects: it does not introduce self-citations or circular derivations, and its descriptive claims are generally traceable to named papers. However, the central claim of being a 'systematic and comprehensive review' is not currently supportable because no selection protocol is given, and one quantitative speedup claim is explicitly admitted to be unverifiable. The survey's significance is therefore conditional on adding a verifiable methodology and removing or qualifying unsupported numbers.

major comments (4)
  1. [§1 and Abstract] The abstract and Section 1 claim a 'systematic and comprehensive review,' but the manuscript provides no search strategy, database list, inclusion/exclusion criteria, time window, or deduplication procedure, and the GitHub repository is an unannotated list of papers. This makes the comprehensiveness claim unfalsifiable: an omission can always be excused as out of scope, and an inclusion as representative. Please either add a methodology subsection that specifies how papers were selected and verified, or adjust the claim to describe a curated survey rather than a systematic and comprehensive one.
  2. [§4.2, DiffusionPipe paragraph] The text states that DiffusionPipe demonstrates 'a 1.41x training speedup' and that data parallelism contributes 'an additional 1.28x acceleration,' then immediately admits that 'the optimization methods for DiffusionPipe were not detailed in the notes.' Reporting an unverifiable quantitative claim as a factual speedup is incompatible with the reliability standard that a systematic review should enforce. Please remove the numbers, verify them against the cited paper, or explicitly label them as unverified claims from the original source.
  3. [Table 2 and §5] Table 2, titled 'Comparison of Optimization Support in Selected Diffusion Model Frameworks and Techniques,' mixes genuinely distinct entities: Flash Attention, xFormers, DeepSpeed, and OneFlow are libraries or frameworks, while DeepCache and TGATE are inference techniques described elsewhere in the survey as caching methods, not standalone frameworks. Several rows also lack citations or version identifiers, e.g., Stable-Fast, Onediff, and xDiT. Because Table 2 is the main synthesis of the frameworks section, this conflation weakens the contribution of §5 and should be corrected by separating frameworks from techniques and providing complete references.
  4. [§1, taxonomy paragraph] The introduction first says the survey is organized into 'three main categories' and then, two sentences later, says 'These four categories cover distinct yet interconnected research topics.' Since the stated taxonomy has three categories, the 'four' appears to be a typographical error, but it appears in the passage that defines the paper's central organizational claim and must be fixed. The same inconsistency should be checked in Figure 1, where 'Frameworks' is rendered as a flat list rather than a category with subdivisions.
minor comments (7)
  1. [§3.3.4] There is a typo in the sentence introducing Consistency Models: 'Song et al. (2023) firtstly introduces' should read 'firstly introduces.'
  2. [Figure 22 and §3.4.1] The name 'Q-Diffsuion' is misspelled; it should be 'Q-Diffusion' to match the cited paper.
  3. [§6] The first future-work paragraph contains 'promissing' and should read 'promising.'
  4. [§4.3 and References] The entries 'Shen et al.' and 'Chen et al.' for MD-DiT and Delta-DiT appear in the text without years and without complete reference entries in the bibliography; please add full citations or remove the incomplete mentions.
  5. [§3.3.4 and References] The early-exit method is cited in the text as 'Moon et al. (2024a)' and the reference list contains both a 2024a arXiv preprint and a 2024b ICML version; please align the citation and reference list.
  6. [§2.2.1] The paper refers to 'Efficient Diffusion (EDM) (Karras et al., 2022),' but Karras et al. is the 'Elucidating the Design Space of Diffusion-Based Generative Models' paper, not a paper titled 'Efficient Diffusion'; please rename or correct this reference to avoid a misleading shorthand.
  7. [§3.2.1] In the LoRA-Composer paragraph, the sentence beginning 'Compared to traditional methods like Mix-of-Show...' is a sentence fragment, and the preceding sentence has an unclear antecedent for 'It'; please revise for readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the paper is an organizational literature review with no derivation, prediction, or fitted input; the only identifiable self-citation is peripheral and not load-bearing.

full rationale

This paper is a survey, so it contains no derivation chain of the kind that can reduce to its own inputs. It asserts a taxonomy and organizes cited works under it, which is a non-empirical organizing choice rather than a prediction or fitted result. The taxonomy's categories are introduced by definition in Section 1 and then applied descriptively; no equation is derived from another in a way that presupposes the conclusion. The only apparent self-citation is Chen et al. (2023b), co-authored by survey author Shoufa Chen, which appears in Section 1 as a general pointer to downstream diffusion applications; that citation is not used to justify the survey's taxonomy, completeness, or any efficiency claim, so it is not load-bearing. The manuscript's acknowledged weakness in Section 4.2, where it reports DiffusionPipe's 1.41x training speedup while admitting 'the optimization methods for DiffusionPipe were not detailed in the notes', is a reliability or transparency concern about a third-party reported number, not a circular argument. Similarly, the inconsistency between 'three main categories' and 'These four categories' in Section 1 is an internal-consistency flaw, not circularity. Because the survey makes no novel empirical predictions and does not fit any parameter to data it later claims to predict, there is no circular step to exhibit, and the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters or invented entities are introduced because this is a review. The load-bearing assumptions are the reliability of cited measurements and the validity of the proposed taxonomy.

assumptions (3)
  • domain assumption The reported efficiency improvements of the surveyed methods (e.g., speedups and FID gains) are accurate as stated in the cited papers.
    The survey does not re-run experiments or independently verify numbers; Section 4.2 explicitly admits that details behind the DiffusionPipe speedup are missing.
  • ad hoc to paper The three-category taxonomy (algorithm-level, system-level, frameworks) is a valid, exhaustive partition of the efficient diffusion model literature.
    Section 1 declares this taxonomy without a formal methodology for paper selection or classification; Figure 1 mixes frameworks with individual techniques, so the partition is not self-evident.
  • domain assumption Efficiency of diffusion models is a well-defined property comparable across different methods without a standardized benchmark.
    The survey treats methods as comparable in efficiency despite the absence of a common evaluation setup.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Diffusion Models: A Survey." pith.science (2026). https://pith.science/paper/BCSS6662

@misc{pith2026250206805,
  author       = {Pith},
  title        = {Pith review of: Efficient Diffusion Models: A Survey},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BCSS6662}},
  note         = {Machine review of arXiv:2502.06805}
}
read the original abstract

Diffusion models have emerged as powerful generative models capable of producing high-quality contents such as images, videos, and audio, demonstrating their potential to revolutionize digital content creation. However, these capabilities come at the cost of their significant computational resources and lengthy generation time, underscoring the critical need to develop efficient techniques for practical deployment. In this survey, we provide a systematic and comprehensive review of research on efficient diffusion models. We organize the literature in a taxonomy consisting of three main categories, covering distinct yet interconnected efficient diffusion model topics from algorithm-level, system-level, and framework perspective, respectively. We have also created a GitHub repository where we organize the papers featured in this survey at https://github.com/AIoT-MLSys-Lab/Efficient-Diffusion-Model-Survey. We hope our survey can serve as a valuable resource to help researchers and practitioners gain a systematic understanding of efficient diffusion model research and inspire them to contribute to this important and exciting field.

Figures

Figures reproduced from arXiv: 2502.06805 by the authors.

Figure 1
Figure 1. Taxonomy of efficient diffusion model literature. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Overview of forward SDE process and reverse SDE process (Song et al., 2020c). [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Summary of efficient training techniques for diffusion models. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (27 more)
Figure 4
Figure 4. Figure 4: Illustration of the rec￾tified flow. Rectified Flow. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]
Figure 5
Figure 5. Figure 5: Illustration of data￾dependent adaptive priors for diffusion processes across differ￾ent modalities. Data-Dependent Adaptive Priors. To enhance the training efficiency of diffusion models and improve the quality of generated samples, data￾dependent adaptive priors can …
Figure 6
Figure 6. Figure 6: The Leapfrog diffusion model (Mao et al., 2023) accelerates inference by using a leapfrog initializer [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Illustration of two categories of noise schedules. [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Summary of efficient fine-tuning techniques for diffusion models. [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Illustration of Hu et al. (2021)’s repa [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Illustration of LCM-LoRA (Luo et al., 2023b). [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Architecture of the Adapter module. Adapters are lightweight modules designed to enable efficient task adaptation by introducing small net￾work layers into pre-trained models, allowing task￾specific feature learning while keeping the original weights frozen. As illust…
Figure 12
Figure 12. Figure 12: Architecture of IP-Adapter (Ye et al., 2023) using a decoupled cross-attention strategy, where [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Illustration of ControlNet. data distributions. Unlike CTRL-Adapter (Lin et al., 2024a), which primarily focuses on attribute control, SimDA is designed to improve the model’s generalization ability, allowing it to generate high-quality content even in unfamiliar data…
Figure 14
Figure 14. Figure 14: Summary of efficient sampling techniques for diffusion models. [PITH_FULL_IMAGE:figures/full_fig_p021_14.png]
Figure 15
Figure 15. Figure 15: Illustration of timestep schedule optimization process. [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Minimizing an upper bound on the Kullback-Leibler divergence (KLUB) between the true and [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: Illustration of the time-dependent exit schedule, where [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: Illustration of the retrieval-based diffusion model. The retriever selects relevant images from a [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]
Figure 19
Figure 19. Figure 19: Illustrations of the knowledge distillation. [PITH_FULL_IMAGE:figures/full_fig_p029_19.png]
Figure 20
Figure 20. Figure 20: Once the student model accurately predicts the teacher’s two-step sampling strategy, it replaces [PITH_FULL_IMAGE:figures/full_fig_p029_20.png]
Figure 21
Figure 21. Figure 21: Illustration as (Poole et al., 2022), it utilizes score distillation sampling. [PITH_FULL_IMAGE:figures/full_fig_p030_21.png]
Figure 22
Figure 22. Figure 22: Summary of compression techniques for DMs. [PITH_FULL_IMAGE:figures/full_fig_p031_22.png]
Figure 23
Figure 23. Figure 23: Traditional PTQ scenarios and Q-Diffusion differ in (a) the creation of calibration datasets and [PITH_FULL_IMAGE:figures/full_fig_p031_23.png]
Figure 24
Figure 24. Figure 24: Pruning evaluates changes in the central tendency and variability to determine the significance [PITH_FULL_IMAGE:figures/full_fig_p032_24.png]
Figure 25
Figure 25. Figure 25: Summary of system-level efficiency optimization techniques for diffusion models. [PITH_FULL_IMAGE:figures/full_fig_p033_25.png]
Figure 26
Figure 26. Figure 26: Illustration of the HybridSA ar￾chitecture from Yang et al. (2023a). In a related effort, Yang et al. (2023a) propose SDA, a low-bit stable diffusion accelerator designed specifically for edge FP￾GAs. Utilizing quantization-aware training and a hybrid sys￾tolic array …
Figure 27
Figure 27. Figure 27: Illustrations of the parallel computing for diffusion models. [PITH_FULL_IMAGE:figures/full_fig_p034_27.png]
Figure 28
Figure 28. Figure 28: Illustrations of the diffusion architecture from (Li et al., 2024a). [PITH_FULL_IMAGE:figures/full_fig_p034_28.png]
Figure 29
Figure 29. Figure 29: Illustrations of the caching system for diffusion models focus on the U-Net block and the Trans [PITH_FULL_IMAGE:figures/full_fig_p035_29.png]
Figure 30
Figure 30. Figure 30: Illustration of the caching system from (Ma et al., 2024b). [PITH_FULL_IMAGE:figures/full_fig_p036_30.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

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

  1. LaTtE-Flow: Layerwise Timestep-Expert Flow-based Transformer

    cs.CV 2025-06 conditional novelty 6.0 of 10

    LaTtE-Flow partitions transformer layers into timestep-specific groups for flow matching, activating only one group per sampling step to speed up image generation in unified multimodal models.

  2. FPSAttention: Training-Aware FP8 and Sparsity Co-Design for Fast Video Diffusion

    cs.CV 2025-06 conditional novelty 6.0 of 10

    FPSAttention co-designs FP8 quantization and sparsity with training, achieving 4.96x end-to-end video generation speedup on Wan2.1 with roughly preserved quality.

  3. Adaptive Visual Autoregressive Acceleration via Dual-Linkage Entropy Analysis

    cs.CV 2026-02 conditional novelty 5.0 of 10

    A training-free entropy-guided token-pruning framework accelerates VAR image generation up to 2.9× with negligible benchmark loss by activating pruning at an adaptive entropy-growth inflection point and adjusting rati...

  4. PWD: Prior-Guided and Wavelet-Enhanced Diffusion Model for Limited-Angle CT

    eess.IV 2025-06 reject novelty 4.0 of 10

    PWD reconstructs limited-angle dental CT in 50 diffusion steps by guiding a DDIM sampler with an FBP prior and fusing wavelet features, reporting higher PSNR/SSIM than baselines, though the reported gain is inconsistent.

Reference graph

Works this paper leans on

84 extracted references · 11 canonical work pages · cited by 4 Pith papers

  1. [4]

    Thibault Castells, Hyoung-Kyu Song, Bo-Kyeong Kim, and Shinkook Choi

    doi: 10.1109/TKDE.2024.3361474. Thibault Castells, Hyoung-Kyu Song, Bo-Kyeong Kim, and Shinkook Choi. Ld-pruner: Efficient pruning of latent diffusion models using task-agnostic insights. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 821–830,

    arXiv 2024

  2. [5]

    A Cheaper and Better Diffusion Language Model with Soft-Masked Noise

    Jiaao Chen, Aston Zhang, Mu Li, Alex Smola, and Diyi Yang. A cheaper and better diffusion language model with soft-masked noise.arXiv preprint arXiv:2304.04746, 2023a. Junsong Chen, Shuchen Xue, Yuyang Zhao, Jincheng Yu, Sayak Paul, Junyu Chen, Han Cai, Enze Xie, and Song Han. Sana-sprint: One-step diffusion with continuous-time consistency distillation.a...

    work page Pith review arXiv

  3. [6]

    An overview of diffusion models: Applications, guided generation, statistical rates and optimization.arXiv preprint arXiv:2404.07771,

    Minshuo Chen, Song Mei, Jianqing Fan, and Mengdi Wang. An overview of diffusion models: Applications, guided generation, statistical rates and optimization.arXiv preprint arXiv:2404.07771,

    arXiv

  4. [7]

    Wavegrad: Estimating gradients for waveform generation.arXiv preprint arXiv:2009.00713,

    Nanxin Chen, Yu Zhang, Heiga Zen, Ron J Weiss, Mohammad Norouzi, and William Chan. Wavegrad: Estimating gradients for waveform generation.arXiv preprint arXiv:2009.00713,

    arXiv 2009

  5. [10]

    Jiwon Choi, Wooyoung Jo, Seongyon Hong, Beomseok Kwon, Wonhoon Park, and Hoi-Jun Yoo. A 28.6 mj/iter stable diffusion processor for text-to-image generation with patch similarity-based sparsity aug- mentation and text-based mixed-precision.arXiv preprint arXiv:2403.04982, 2024a. Joo Young Choi, Jaesung R Park, Inkyu Park, Jaewoong Cho, Albert No, and Erne...

    work page Pith review arXiv

  6. [12]

    Dale Decatur, Itai Lang, Kfir Aberman, and Rana Hanocka

    URL https://arxiv.org/abs/2410.00083. Dale Decatur, Itai Lang, Kfir Aberman, and Rana Hanocka. 3d paintbrush: Local stylization of 3d shapes with cascaded score distillation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4473–4483,

    arXiv

  7. [13]

    Efficient- vdit: Efficient video diffusion transformers with attention tile.arXiv preprint arXiv:2502.06155,

    Hangliang Ding, Dacheng Li, Runlong Su, Peiyuan Zhang, Zhijie Deng, Ion Stoica, and Hao Zhang. Efficient- vdit: Efficient video diffusion transformers with attention tile.arXiv preprint arXiv:2502.06155,

    arXiv

  8. [14]

    Score-based generative modeling with critically-damped langevin diffusion

    Tim Dockhorn, Arash Vahdat, and Karsten Kreis. Score-based generative modeling with critically-damped langevin diffusion. arXiv preprint arXiv:2112.07068,

    arXiv

Show all 84 references
  1. [16]

    Learned step size quantization.arXiv preprint arXiv:1902.08153,

    Steven K Esser, Jeffrey L McKinstry, Deepika Bablani, Rathinakumar Appuswamy, and Dharmendra S Modha. Learned step size quantization.arXiv preprint arXiv:1902.08153,

    1902 arXiv

  2. [17]

    A unified sequence parallelism approach for long context generative ai

    Jiarui Fang and Shangchun Zhao. A unified sequence parallelism approach for long context generative ai. arXiv preprint arXiv:2405.07719,

    arXiv

  3. [18]

    xdit: an inference engine for diffusion transformers (dits) with massive parallelism.arXiv preprint arXiv:2411.01738, 2024a

    Jiarui Fang, Jinzhe Pan, Xibo Sun, Aoyu Li, and Jiannan Wang. xdit: an inference engine for diffusion transformers (dits) with massive parallelism.arXiv preprint arXiv:2411.01738, 2024a. Jiarui Fang, Jinzhe Pan, Jiannan Wang, Aoyu Li, and Xibo Sun. Pipefusion: Patch-level pipe...

    arXiv

  4. [19]

    Concept sliders: Lora adaptors for precise control in diffusion models.arXiv preprint arXiv:2311.12092,

    Rohit Gandikota, Joanna Materzynska, Tingrui Zhou, Antonio Torralba, and David Bau. Concept sliders: Lora adaptors for precise control in diffusion models.arXiv preprint arXiv:2311.12092,

    arXiv

  5. [20]

    Shansan Gong, Mukai Li, Jiangtao Feng, Zhiyong Wu, and LingPeng Kong

    URLhttps://diffusionflow.github .io/. Shansan Gong, Mukai Li, Jiangtao Feng, Zhiyong Wu, and LingPeng Kong. Diffuseq: Sequence to sequence text generation with diffusion models.arXiv preprint arXiv:2210.08933,

    arXiv

  6. [21]

    Improved noise schedule for diffusion training

    Tiankai Hang and Shuyang Gu. Improved noise schedule for diffusion training. arXiv preprint arXiv:2407.03297,

    arXiv

  7. [22]

    Efficientdm: Efficient quantization-aware fine-tuning of low-bit diffusion models.arXiv preprint arXiv:2310.03270,

    Yefei He, Jing Liu, Weijia Wu, Hong Zhou, and Bohan Zhuang. Efficientdm: Efficient quantization-aware fine-tuning of low-bit diffusion models.arXiv preprint arXiv:2310.03270,

    arXiv

  8. [23]

    Latent video diffusion models for high-fidelity long video generation.arXiv preprint arXiv:2211.13221,

    Yingqing He, Tianyu Yang, Yong Zhang, Ying Shan, and Qifeng Chen. Latent video diffusion models for high-fidelity long video generation.arXiv preprint arXiv:2211.13221,

    arXiv

  9. [24]

    Iterativeα-(de) blending: A minimalist deterministic diffusion model

    Eric Heitz, Laurent Belcour, and Thomas Chambon. Iterativeα-(de) blending: A minimalist deterministic diffusion model. InACM SIGGRAPH 2023 Conference Proceedings, pp. 1–8,

    2023

  10. [25]

    Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531,

    41 Published in Transactions on Machine Learning Research (05/2025) Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network.arXiv preprint arXiv:1503.02531,

    2025 arXiv

  11. [27]

    Latent diffusion models for controllable rna sequence generation.arXiv preprint arXiv:2409.09828,

    Kaixuan Huang, Yukang Yang, Kaidi Fu, Yanyi Chu, Le Cong, and Mengdi Wang. Latent diffusion models for controllable rna sequence generation.arXiv preprint arXiv:2409.09828,

    arXiv

  12. [28]

    Gotta go fast when generating data with score-based models.arXiv preprint arXiv:2105.14080,

    Alexia Jolicoeur-Martineau, Ke Li, Rémi Piché-Taillefer, Tal Kachman, and Ioannis Mitliagkas. Gotta go fast when generating data with score-based models.arXiv preprint arXiv:2105.14080,

    arXiv

  13. [29]

    doi: https://doi.org/10.1016/j.media.2023.102846

    ISSN 1361-8415. doi: https://doi.org/10.1016/j.media.2023.102846. URL https://www.sciencedirect.com/science/article/pii/S1361841523001068. Beomsu Kim and Jong Chul Ye. Denoising mcmc for accelerating diffusion-based generative models.arXiv preprint arXiv:2209.14593,

    2023

  14. [30]

    Layermerge: Neural network depth compression through layer pruning and merging.arXiv preprint arXiv:2406.12837, 2024a

    Jinuk Kim, Marwa El Halabi, Mingi Ji, and Hyun Oh Song. Layermerge: Neural network depth compression through layer pruning and merging.arXiv preprint arXiv:2406.12837, 2024a. Yulhwa Kim, Dongwon Jo, Hyesung Jeon, Taesu Kim, Daehyun Ahn, Hyungjun Kim, et al. Leveraging early- s...

    arXiv

  15. [32]

    On fast sampling of diffusion probabilistic models

    Zhifeng Kong and Wei Ping. On fast sampling of diffusion probabilistic models. arXiv preprint arXiv:2106.00132,

    arXiv

  16. [33]

    Diffwave: A versatile diffusion model for audio synthesis.arXiv preprint arXiv:2009.09761,

    42 Published in Transactions on Machine Learning Research (05/2025) Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. Diffwave: A versatile diffusion model for audio synthesis.arXiv preprint arXiv:2009.09761,

    2025 arXiv

  17. [34]

    Improving the training of rectified flows

    Sangyun Lee, Zinan Lin, and Giulia Fanti. Improving the training of rectified flows. arXiv preprint arXiv:2405.20320, 2024a. Taegyeong Lee, Soyeong Kwon, and Taehwan Kim. Grid diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF Conference on Computer ...

    arXiv

  18. [35]

    Distrifusion: Distributed parallel inference for high-resolution diffusion models

    Muyang Li, Tianle Cai, Jiaxin Cao, Qinsheng Zhang, Han Cai, Junjie Bai, Yangqing Jia, Kai Li, and Song Han. Distrifusion: Distributed parallel inference for high-resolution diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p...

    arXiv

  19. [36]

    Dual3d: Efficient and consistent text-to-3d generation with dual-mode multi-view latent diffusion

    Xinyang Li, Zhangyu Lai, Linning Xu, Jianfei Guo, Liujuan Cao, Shengchuan Zhang, Bo Dai, and Rongrong Ji. Dual3d: Efficient and consistent text-to-3d generation with dual-mode multi-view latent diffusion. arXiv preprint arXiv:2405.09874, 2024c. Xiuyu Li, Yijiang Liu, Long Lian...

    arXiv

  20. [37]

    Immiscible diffusion: Accelerating diffusion training with noise assignment.arXiv preprint arXiv:2406.12303, 2024d

    Yiheng Li, Heyang Jiang, Akio Kodaira, Masayoshi Tomizuka, Kurt Keutzer, and Chenfeng Xu. Immiscible diffusion: Accelerating diffusion training with noise assignment.arXiv preprint arXiv:2406.12303, 2024d. Yuhang Li, Ruihao Gong, Xu Tan, Yang Yang, Peng Hu, Qi Zhang, Fengwei Y...

    arXiv

  21. [38]

    Ctrl-adapter: An efficient and versatile framework for adapting diverse controls to any diffusion model.arXiv preprint arXiv:2404.09967, 2024a

    Han Lin, Jaemin Cho, Abhay Zala, and Mohit Bansal. Ctrl-adapter: An efficient and versatile framework for adapting diverse controls to any diffusion model.arXiv preprint arXiv:2404.09967, 2024a. Jianghao Lin, Jiaqi Liu, Jiachen Zhu, Yunjia Xi, Chengkai Liu, Yangtian Zhang, Yon...

    2025 arXiv

  22. [39]

    Audioldm: Text-to-audio generation with latent diffusion models.arXiv preprint arXiv:2301.12503, 2023a

    Haohe Liu, Zehua Chen, Yi Yuan, Xinhao Mei, Xubo Liu, Danilo Mandic, Wenwu Wang, and Mark D Plumb- ley. Audioldm: Text-to-audio generation with latent diffusion models.arXiv preprint arXiv:2301.12503, 2023a. Haozhe Liu, Wentian Zhang, Jinheng Xie, Francesco Faccio, Mengmeng Xu...

    arXiv

  23. [40]

    Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022b

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022b. Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, et al. Instaflow: One step is enough for high-quality diffus...

    arXiv

  24. [41]

    Simplifying, stabilizing and scaling continuous-time consistency models.arXiv preprint arXiv:2410.11081,

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

    arXiv

  25. [42]

    Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models.arXiv preprint arXiv:2211.01095,

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models.arXiv preprint arXiv:2211.01095,

    arXiv

  26. [43]

    Knowledge distillation in iterative generative models for improved sampling speed

    Eric Luhman and Troy Luhman. Knowledge distillation in iterative generative models for improved sampling speed. arXiv preprint arXiv:2101.02388,

    arXiv

  27. [44]

    Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023a

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high-resolution images with few-step inference.arXiv preprint arXiv:2310.04378, 2023a. Simian Luo, Yiqin Tan, Suraj Patil, Daniel Gu, Patrick von Platen, Apolinário Passos, Lon...

    arXiv

  28. [45]

    Learning-to-cache: Accelerating diffusion transformer via layer caching.arXiv preprint arXiv:2406.01733, 2024a

    Xinyin Ma, Gongfan Fang, Michael Bi Mi, and Xinchao Wang. Learning-to-cache: Accelerating diffusion transformer via layer caching.arXiv preprint arXiv:2406.01733, 2024a. Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free. In Proceeding...

    2025 arXiv

  29. [46]

    Video diffusion models: A survey.arXiv preprint arXiv:2405.03150,

    Andrew Melnik, Michal Ljubljanac, Cong Lu, Qi Yan, Weiming Ren, and Helge Ritter. Video diffusion models: A survey.arXiv preprint arXiv:2405.03150,

    arXiv

  30. [47]

    Efficient 3d shape generation via diffusion mamba with bidirectional ssms.arXiv preprint arXiv:2406.05038,

    Shentong Mo. Efficient 3d shape generation via diffusion mamba with bidirectional ssms.arXiv preprint arXiv:2406.05038,

    arXiv

  31. [48]

    A simple early exiting framework for accelerated sampling in diffusion models

    Taehong Moon, Moonseok Choi, EungGu Yun, Jongmin Yoon, Gayoung Lee, Jaewoong Cho, and Juho Lee. A simple early exiting framework for accelerated sampling in diffusion models. arXiv preprint arXiv:2408.05927, 2024a. Taehong Moon, Moonseok Choi, EungGu Yun, Jongmin Yoon, Gayoung...

    arXiv

  32. [49]

    The blessing of randomness: Sde beats ode in general diffusion-based image editing.arXiv preprint arXiv:2311.01410,

    Shen Nie, Hanzhong Allan Guo, Cheng Lu, Yuhao Zhou, Chenyu Zheng, and Chongxuan Li. The blessing of randomness: Sde beats ode in general diffusion-based image editing.arXiv preprint arXiv:2311.01410,

    arXiv

  33. [50]

    Ed-nerf: Efficient text-guided editing of 3d scene using latent space nerf.arXiv preprint arXiv:2310.02712,

    Jangho Park, Gihyun Kwon, and Jong Chul Ye. Ed-nerf: Efficient text-guided editing of 3d scene using latent space nerf.arXiv preprint arXiv:2310.02712,

    arXiv

  34. [51]

    Non-denoising forward-time diffusions.arXiv preprint arXiv:2312.14589,

    Stefano Peluchetti. Non-denoising forward-time diffusions.arXiv preprint arXiv:2312.14589,

    arXiv

  35. [52]

    Controlnext: Powerful and efficient control for image and video generation.arXiv preprint arXiv:2408.06070, 2024a

    Bohao Peng, Jian Wang, Yuechen Zhang, Wenbo Li, Ming-Chang Yang, and Jiaya Jia. Controlnext: Powerful and efficient control for image and video generation.arXiv preprint arXiv:2408.06070, 2024a. Mingxing Peng, Kehua Chen, Xusen Guo, Qiming Zhang, Hongliang Lu, Hui Zhong, Di Ch...

    arXiv

  36. [53]

    Barron, and Ben Mildenhall

    45 Published in Transactions on Machine Learning Research (05/2025) Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. arXiv,

    2025

  37. [54]

    Unicontrol: A unified diffusion model for controllable visual gener- ation in the wild.arXiv preprint arXiv:2305.11147,

    Can Qin, Shu Zhang, Ning Yu, Yihao Feng, Xinyi Yang, Yingbo Zhou, Huan Wang, Juan Carlos Niebles, Caiming Xiong, Silvio Savarese, et al. Unicontrol: A unified diffusion model for controllable visual gener- ation in the wild.arXiv preprint arXiv:2305.11147,

    arXiv

  38. [55]

    Wasserstein barycenter and its application to texture mixing

    Julien Rabin, Gabriel Peyré, Julie Delon, and Marc Bernot. Wasserstein barycenter and its application to texture mixing. In Scale Space and Variational Methods in Computer Vision: Third International Conference, SSVM 2011, Ein-Gedi, Israel, May 29–June 2, 2011, Revised Selecte...

    2011

  39. [56]

    Hierarchical text-conditional image generation with clip latents.arXiv preprint arXiv:2204.06125,

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents.arXiv preprint arXiv:2204.06125,

    arXiv

  40. [58]

    Align your steps: Optimizing sampling schedules in diffusion models

    Amirmojtaba Sabour, Sanja Fidler, and Karsten Kreis. Align your steps: Optimizing sampling schedules in diffusion models. arXiv preprint arXiv:2404.14507,

    arXiv

  41. [59]

    Consistent accelerated inference via confident adaptive transformers.arXiv preprint arXiv:2104.08803,

    Tal Schuster, Adam Fisch, Tommi Jaakkola, and Regina Barzilay. Consistent accelerated inference via confident adaptive transformers.arXiv preprint arXiv:2104.08803,

    arXiv

  42. [60]

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

    Pratheba Selvaraju, Tianyu Ding, Tianyi Chen, Ilya Zharkov, and Luming Liang. Fora: Fast-forward caching in diffusion transformer acceleration.arXiv preprint arXiv:2407.01425,

    arXiv

  43. [61]

    Post-training quantization on diffusion models

    Yuzhang Shang, Zhihang Yuan, Bin Xie, Bingzhe Wu, and Yan Yan. Post-training quantization on diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 1972–1981,

    1972

  44. [62]

    Parallel sampling of diffusion models

    46 Published in Transactions on Machine Learning Research (05/2025) Andy Shih, Suneel Belkhale, Stefano Ermon, Dorsa Sadigh, and Nima Anari. Parallel sampling of diffusion models. Advances in Neural Information Processing Systems, 36,

    2025

  45. [65]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. arXiv preprint arXiv:2303.01469,

    arXiv

  46. [66]

    Unveiling redundancy in diffusion transformers (dits): A systematic study.arXiv preprint arXiv:2411.13588,

    Xibo Sun, Jiarui Fang, Aoyu Li, and Jinzhe Pan. Unveiling redundancy in diffusion transformers (dits): A systematic study.arXiv preprint arXiv:2411.13588,

    arXiv

  47. [67]

    Deediff: Dynamic uncertainty-aware early exiting for accelerating diffusion model generation

    Shengkun Tang, Yaqing Wang, Caiwen Ding, Yi Liang, Yao Li, and Dongkuan Xu. Deediff: Dynamic uncertainty-aware early exiting for accelerating diffusion model generation. 2023a. Zhicong Tang, Shuyang Gu, Chunyu Wang, Ting Zhang, Jianmin Bao, Dong Chen, and Baining Guo. Volumedi...

    arXiv

  48. [68]

    Anwaar Ulhaq, Naveed Akhtar, and Ganna Pogrebna

    URLhttps://arxiv.org/abs/2405.01248. Anwaar Ulhaq, Naveed Akhtar, and Ganna Pogrebna. Efficient diffusion models for vision: A survey.arXiv preprint arXiv:2210.09292,

    arXiv

  49. [69]

    Anderson acceleration for fixed-point iterations.SIAM Journal on Numerical Analysis, 49(4):1715–1735,

    47 Published in Transactions on Machine Learning Research (05/2025) Homer F Walker and Peng Ni. Anderson acceleration for fixed-point iterations.SIAM Journal on Numerical Analysis, 49(4):1715–1735,

    2025

  50. [71]

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

    Zhengyi Wang, Cheng Lu, Yikai Wang, Fan Bao, Chongxuan Li, Hang Su, and Jun Zhu. Prolificdreamer: High-fidelity and diverse text-to-3d generation with variational score distillation. Advances in Neural Information Processing Systems, 36, 2024b. Daniel Watson, Jonathan Ho, Moha...

    arXiv

  51. [72]

    Training-free and adaptive sparse attention for efficient long video generation.arXiv preprint arXiv:2502.21079,

    Yifei Xia, Suhan Ling, Fangcheng Fu, Yujie Wang, Huixia Li, Xuefeng Xiao, and Bin Cui. Training-free and adaptive sparse attention for efficient long video generation.arXiv preprint arXiv:2502.21079,

    arXiv

  52. [73]

    Versatile diffusion: Text, images and variations all in one diffusion model

    Xingqian Xu, Zhangyang Wang, Gong Zhang, Kai Wang, and Humphrey Shi. Versatile diffusion: Text, images and variations all in one diffusion model. InProceedings of the IEEE/CVF International Conference on Computer Vision, pp. 7754–7765, 2023a. YilunXu, MingyangDeng, XiangCheng,...

    arXiv

  53. [74]

    Sda: Low-bit stable diffusion acceleration on edge fpgas

    Geng Yang, Yanyue Xie, Zhong Jia Xue, Sung-En Chang, Yanyu Li, Peiyan Dong, Jie Lei, Weiying Xie, Yanzhi Wang, Xue Lin, et al. Sda: Low-bit stable diffusion acceleration on edge fpgas. 2023a. Ling Yang, Zhilong Zhang, Yang Song, Shenda Hong, Runsheng Xu, Yue Zhao, Wentao Zhang...

    2025 arXiv

  54. [75]

    Reconstruction vs

    Jingfeng Yao, Bin Yang, and Xinggang Wang. Reconstruction vs. generation: Taming optimization dilemma in latent diffusion models.arXiv preprint arXiv:2501.01423,

    arXiv

  55. [76]

    Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models.arXiv preprint arXiv:2308.06721,

    Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models.arXiv preprint arXiv:2308.06721,

    arXiv

  56. [77]

    Nuwa-xl: Diffusion over diffusion for extremely long video generation

    Shengming Yin, Chenfei Wu, Huan Yang, Jianfeng Wang, Xiaodong Wang, Minheng Ni, Zhengyuan Yang, Linjie Li, Shuguang Liu, Fan Yang, et al. Nuwa-xl: Diffusion over diffusion for extremely long video generation. arXiv preprint arXiv:2303.12346,

    arXiv

  57. [78]

    Representation alignment for generation: Training diffusion transformers is easier than you think

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffusion transformers is easier than you think. arXiv preprint arXiv:2410.06940, 2024a. YonghaoYu, ShunanZhu, HuaiQin, andHa...

    arXiv

  58. [79]

    Oneflow: Redesign the distributed deep learning framework from scratch.arXiv preprint arXiv:2110.15032,

    Jinhui Yuan, Xinqi Li, Cheng Cheng, Juncheng Liu, Ran Guo, Shenghang Cai, Chi Yao, Fei Yang, Xiaodong Yi, Chuan Wu, et al. Oneflow: Redesign the distributed deep learning framework from scratch.arXiv preprint arXiv:2110.15032,

    arXiv

  59. [80]

    Diffusion models with deterministic normalizing flow priors

    Mohsen Zand, Ali Etemad, and Michael Greenspan. Diffusion models with deterministic normalizing flow priors. arXiv preprint arXiv:2309.01274,

    arXiv

  60. [81]

    Laptop-diff: Layer pruning and normalized distillation for compressing diffusion models.arXiv preprint arXiv:2404.11098, 2024a

    Dingkun Zhang, Sijia Li, Chen Chen, Qingsong Xie, and Haonan Lu. Laptop-diff: Layer pruning and normalized distillation for compressing diffusion models.arXiv preprint arXiv:2404.11098, 2024a. Hengrui Zhang, Jiani Zhang, Balasubramaniam Srinivasan, Zhengyuan Shen, Xiao Qin, Ch...

    arXiv

  61. [82]

    Fast sampling of diffusion models with exponential integrator.arXiv preprint arXiv:2204.13902,

    Qinsheng Zhang and Yongxin Chen. Fast sampling of diffusion models with exponential integrator.arXiv preprint arXiv:2204.13902,

    arXiv

  62. [83]

    Magicvideo: Efficient video generation with latent diffusion models.arXiv preprint arXiv:2211.11018,

    Daquan Zhou, Weimin Wang, Hanshu Yan, Weiwei Lv, Yizhe Zhu, and Jiashi Feng. Magicvideo: Efficient video generation with latent diffusion models.arXiv preprint arXiv:2211.11018,

    arXiv

  63. [84]

    Emdm: Efficient motion diffusion model for fast, high-quality motion generation

    Wenyang Zhou, Zhiyang Dou, Zeyu Cao, Zhouyingcheng Liao, Jingbo Wang, Wenjia Wang, Yuan Liu, Taku Komura, Wenping Wang, and Lingjie Liu. Emdm: Efficient motion diffusion model for fast, high-quality motion generation. Yuanzhi Zhu, Xingchao Liu, and Qiang Liu. Slimflow: Trainin...

    arXiv

  64. [2009]

    L3dg: Latent 3d gaussian diffusion

    Barbara Roessle, Norman Müller, Lorenzo Porzi, Samuel Rota Bulò, Peter Kontschieder, Angela Dai, and Matthias Nießner. L3dg: Latent 3d gaussian diffusion. InSIGGRAPH Asia 2024 Conference Papers, pp. 1–11,

    2024

  65. [2010]

    Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685,

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models.arXiv preprint arXiv:2106.09685,

    arXiv

  66. [2011]

    Pipefusion: Displaced patch pipeline parallelism for inference of diffusion transformer models.arXiv preprint arXiv:2405.14430, 2024a

    Jiannan Wang, Jiarui Fang, Aoyu Li, and PengCheng Yang. Pipefusion: Displaced patch pipeline parallelism for inference of diffusion transformer models.arXiv preprint arXiv:2405.14430, 2024a. Xiang Wang, Shiwei Zhang, Han Zhang, Yu Liu, Yingya Zhang, Changxin Gao, and Nong Sang...

    arXiv

  67. [2013]

    Streamdiffusion: A pipeline-level solution for real-time interactive generation.arXiv preprint arXiv:2312.12491,

    Akio Kodaira, Chenfeng Xu, Toshiki Hazama, Takanori Yoshimoto, Kohei Ohno, Shogo Mitsuhori, Soichi Sugano, Hanying Cho, Zhijian Liu, and Kurt Keutzer. Streamdiffusion: A pipeline-level solution for real-time interactive generation.arXiv preprint arXiv:2312.12491,

    arXiv

  68. [2015]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502, 2020a. Yang Song and Prafulla Dhariwal. Improved techniques for training consistency models.arXiv preprint arXiv:2310.14189,

    2010 arXiv

  69. [2018]

    Diffusiondet: Diffusion model for object detection

    Shoufa Chen, Peize Sun, Yibing Song, and Ping Luo. Diffusiondet: Diffusion model for object detection. In Proceedings of the IEEE/CVF international conference on computer vision, pp. 19830–19843, 2023b. Ting Chen, Ruixiang Zhang, and Geoffrey Hinton. Analog bits: Generating di...

    arXiv

  70. [2019]

    Sliced score matching: A scalable approach to density and score estimation

    Yang Song, Sahaj Garg, Jiaxin Shi, and Stefano Ermon. Sliced score matching: A scalable approach to density and score estimation. InUncertainty in Artificial Intelligence, pp. 574–584. PMLR, 2020b. Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Er...

    2011 arXiv

  71. [2020]

    39 Published in Transactions on Machine Learning Research (05/2025) Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud

    Pengtao Chen, Mingzhu Shen, Peng Ye, Jianjian Cao, Chongjun Tu, Christos-Savvas Bouganis, Yiren Zhao, and Tao Chen.δ-dit: Accelerating diffusion transformers without training via denoising property align- ment. 39 Published in Transactions on Machine Learning Research (05/2025...

    2025

  72. [2021]

    Multi-scale latent point consistency models for 3d shape generation

    40 Published in Transactions on Machine Learning Research (05/2025) Bi’an Du, Wei Hu, and Renjie Liao. Multi-scale latent point consistency models for 3d shape generation. arXiv preprint arXiv:2412.19413,

    2025

  73. [2022]

    Diffedit: Diffusion-basedsemantic image editing with mask guidance.arXiv preprint arXiv:2210.11427,

    GuillaumeCouairon, JakobVerbeek, HolgerSchwenk, andMatthieuCord. Diffedit: Diffusion-basedsemantic image editing with mask guidance.arXiv preprint arXiv:2210.11427,

    arXiv

  74. [2023]

    Semi-parametric neural image synthesis

    Andreas Blattmann, Robin Rombach, Kaan Oktay, Jonas Müller, and Björn Ommer. Semi-parametric neural image synthesis. arXiv preprint arXiv:2204.11824,

    arXiv

  75. [2024]

    Stochastic interpolants: A unifying frame- work for flows and diffusions.arXiv preprint arXiv:2303.08797,

    Michael S Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying frame- work for flows and diffusions.arXiv preprint arXiv:2303.08797,

    arXiv

  76. [2025]

    Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu

    URL https://arxiv.org/abs/2503.09573. Fan Bao, Shen Nie, Kaiwen Xue, Yue Cao, Chongxuan Li, Hang Su, and Jun Zhu. All are worth words: A vit backbone for diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 22669–22679,

    arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.