Pith. sign in

REVIEW 3 major objections 3 minor 57 references

Inconsistencies In Consistency Models: Better ODE Solving Does Not Imply Better Samples

T0 review · 3 major / 3 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Directly training consistency models to minimize ODE solving error produces worse sample quality than the standard weakly-supervised objective.

desk verdict A well-controlled empirical probe showing that better ODE-solver matching can hurt sample quality, but the headline needs an independent ODE-error check before it fully lands. read the letter →

arxiv 2411.08954 v2 pith:4DFWFHTR submitted 2024-11-13 cs.LG cs.AI

Noël Vouitsis , Rasa Hosseinzadeh , Brendan Leigh Ross , Valentin Villecroze , Satya Krishna Gorti , Jesse C. Cresswell , Gabriel Loaiza-Ganem This is my paper · ORCID
classification cs.LGcs.AI
keywords consistencymodelsdiffusiondistillationprobabilityflowODEsolvingerrorfew-stepgenerationsamplequalityDirectlatent
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

Consistency models distill a diffusion model by training a network to mimic a numerical solver of the probability-flow ODE, with the implicit assumption that more faithful ODE solving means better samples. The paper tests this assumption directly by introducing Direct Consistency Models, which replace the standard self-consistency loss with a loss that explicitly minimizes the distance to the solver's output at every point along the trajectory. In controlled experiments distilling SDXL with three different solvers and across ablations of discretization and guidance scale, Direct CMs achieve lower ODE solving error $E$ but noticeably worse image quality on every image metric. The authors conclude that the success of consistency models cannot be attributed to their ODE-solving fidelity, and that other factors must drive sample quality.

What carries the argument

The load-bearing object is the Direct CM objective (Equation 5), $L_{\mathrm{CD}}^{\mathrm{Direct}} = \mathbb{E}[\lambda(t_n)\, d(f_\theta(x_{t_n}, t_n), f_{\mathrm{solver}}(x_{t_n}, t_n, 0))]$, which replaces the weak self-consistency loss (Equation 4) with direct, per-point supervision from the discrete teacher solver. The companion metric $E$ (Equation 6) measures exactly the expected distance this loss minimizes, so the paper's ODE-solving comparison is, by construction, a comparison of how well each model matches $f_{\mathrm{solver}}$ from the same initial noise. The controlled setup isolates the loss as the only changed component, with identical teacher (SDXL), LoRA parameterization, seeds, and evaluation protocol.

What would settle it

Compute $E$ for both models against a reference solution obtained from a much finer discretization or a high-order solver with thousands of steps, and check whether Direct CMs still have lower error to that reference while producing worse images; if the error ordering reverses, the central claim reduces to a statement about matching a fixed training target.

Watch

Extended reading notes

Core claim

The central discovery is a counterexample to the assumption that better PF ODE solving entails better sample quality. The paper constructs Direct CMs, trained with Equation 5, which supervises $f_\theta(x_{t_n}, t_n)$ to match $f_{\mathrm{solver}}(x_{t_n}, t_n, 0)$ directly, so that minimizing the objective is essentially minimizing the measured ODE error $E$ of Equation 6. Across DDIM, Euler, and Heun solvers, and across $N \in \{25, 50, 100, 200\}$ and $\omega \in \{1, 4, 8, 11\}$, Direct CMs consistently reduce $E$ relative to standard CMs yet produce worse FID, FD-DINO, CLIP, and aesthetic scores, with visibly noisier images. The paper argues this falsifies the implicit assumption behind CM-style distillation and calls for investigation of confounding factors that make CMs work.

Load-bearing premise

The argument depends on treating the numerical solver used for supervision as a faithful stand-in for the true probability-flow solution; if that solver is far from the true solution, the result only shows that matching the training target better does not improve samples.

Editorial extensions

If this is right

  • If valid, the counterexample refutes the notion that better PF ODE solving is the mechanism by which consistency models obtain high sample quality.
  • Improvements attributed to follow-up CM variants that add trajectory splitting or bidirectional solving should not be credited to better ODE fidelity without independent evidence.
  • ODE-solving fidelity alone is not a reliable proxy for evaluating distillation quality, so sample-based metrics remain necessary.
  • The Direct CM design provides a reusable probe: any distillation method can be tested against $f_{\mathrm{solver}}$ with Equation 6 to separate ODE accuracy from generation quality.

Reading between the lines

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

  • The paper's own caveats suggest a stronger reading: because $E$ is measured against the discrete solver, the result may show that matching the training target better does not improve samples, rather than that matching the true solution does not; testing against a much finer reference solver would settle which statement holds.
  • The observed gap hints that the self-consistency loss acts as an implicit regularizer or denoiser, and removing it removes a beneficial inductive bias; one testable extension is to add the self-consistency term back to the Direct CM loss and see whether sample quality recovers while $E$ stays low.
  • A practical consequence the authors do not draw is that ODE-faithful distillation may be the wrong target for few-step generation, and methods that add distribution-matching or adversarial losses may be aligning with the right objective for reasons unrelated to the ODE.
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 / 3 minor

Summary. The paper introduces "Direct CMs," a variant of consistency-model distillation trained with Eq. (5) to directly minimize the distance between the student's one-step prediction and the numerical solver's output along the probability-flow ODE. Using SDXL as the teacher and LoRA fine-tuning, the authors compare Direct CMs against standard CMs across three ODE solvers (DDIM, Euler, Heun), for one, two, and four sampling steps, and across ablations over discretization intervals N and guidance scales ω. The reported results show that Direct CMs achieve lower ODE-solving error E (Eq. 6) but worse image-quality metrics (FID, FD-DINO, CLIP, aesthetic score). The paper concludes that better PF ODE solving does not imply better sample quality, and questions the theoretical motivation of consistency models.

Significance. If the dissociation between ODE-solving fidelity and sample quality were firmly established, this would be a valuable counterexample for the distillation community, with implications for consistency models and trajectory-based methods more broadly. The study has strengths: it is a controlled comparison where only the loss is changed, it uses a large-scale teacher, it releases code, and it includes sweeps over solver type, discretization, and guidance scale. However, the central ODE-error measure E is the final-timestep term of the Direct CM training objective, so the headline claim is not supported by an independent measurement of ODE-solving quality. As it stands, the experiments convincingly show that better matching the training target does not imply better samples, which is weaker than the paper's stated conclusion.

major comments (3)
  1. [Section 3, Eqs. (5) and (6)] The definition of E in Eq. (6) is the n = N term of the Direct CM training objective in Eq. (5). Since Direct CMs are trained to minimize Eq. (5) for all n, including n = N, observing E_DirectCM < E_CM is substantially built into the training procedure rather than being an independent measurement of ODE-solving error. The paper never verifies that the numerical solver fsolver with N = 50–200 and guidance scales up to 11 is close to the exact solution f^s of the empirical PF ODE in Eq. (3), for example by comparing against a high-accuracy reference solution with much larger N or adaptive step size. Without such verification, the results support the weaker statement that directly minimizing the solver-distance objective does not imply better samples, not the stronger claim in the abstract and Section 4 that better ODE solving does not imply better samples.
  2. [Section 4, Table 1 and Figures 3–4] All reported numbers are point estimates from a single training run of 250 LoRA steps, with no error bars, multiple seeds, or significance tests. The image-quality gap is large, but the ODE-error differences are modest (E of 0.23–0.25 for Direct CMs versus 0.29–0.30 for CMs), and without run-to-run variability estimates the claim that Direct CMs are "consistently" better at ODE solving is not statistically supported. At minimum, the paper should report confidence intervals or repeated training runs for the main comparisons.
  3. [Appendix A.2, Table 3] The claim that CMs produce higher-quality images than Direct CMs is not uniformly true in the reported multi-step results. At four steps, Direct CMs outperform CMs on FD-DINO for both DDIM (155.1 versus 159.8) and Heun (150.6 versus 159.4), and on CLIP score for both (0.28 versus 0.27). The paper attributes these exceptions to imperfections in automated metrics, but this is a post hoc explanation. The conclusions should either be scoped to single-step generation or supported by additional analysis showing that the metric exceptions do not reflect genuine sample-quality differences.
minor comments (3)
  1. [Author list] The first author's name contains an erroneous space: "Noël V ouitsis" should read "Noël Vouitsis."
  2. [Section 2, Eq. (4)] The notation "n∼U J1,N K" is malformed; it should be written as n ∼ Uniform({1, ..., N}) or equivalent.
  3. [Appendix A.1] The paper states that metric and loss curves stabilized around 250 training steps, but no learning curves are shown. Since Direct CMs optimize a different objective, a brief sensitivity check over training length, LoRA rank, or loss weighting λ(t) would make the controlled comparison more persuasive.

Circularity Check

1 steps flagged · score 6.0 of 10

Partially circular: the ODE-solving error E (Eq. 6) is the n=N term of the Direct CM training objective (Eq. 5), so Direct CMs' lower E is built into the fit rather than an independent ODE-fidelity benchmark; only the image-quality gap is independently established.

  1. self definitional [Section 3, Equations (5)-(6); operationalized in Section 4, Table 1 and Figures 3-4]
    "LDirect CD := Ex0∼p0,n∼U J1,N K,xtn ∼ptn |0(·|x0)[λ(tn)d(fθ(xtn,tn), fsolver(xtn,tn,0))], (5) ... We define the ODE solving error E as the expected distance between the ODE solver’s solution and the CM’s prediction with the same initial noise, i.e., E := ExT ∼bpT [d(fθ(xT,T), fsolver(xT,T,0))]. (6)"

    Equation (6) is, up to replacing p_T by the approximate bp_T, the n=N term of Equation (5), the Direct CM training objective. Direct CMs minimize d(fθ(xtn,tn), fsolver(xtn,tn,0)) at every timestep including T, so E_DirectCM < E_CM is forced by fitting to fsolver rather than measured independently. The paper never quantifies the gap between fsolver and the exact empirical PF ODE solution f^s of Equation (3); explanation (ii) concerns f^s ≠ f^*, not fsolver ≈ f^s. So 'better ODE solving' may mean only 'better matching of the numerical solver used as supervision.' Image-quality metrics are independent, but the ODE-solving premise is the training target renamed as an error metric.

full rationale

The paper's empirical claim has two halves. The image-quality half is a clean controlled experiment: the only training change is the loss (Equations 4 vs 5), and FID, FD-DINO, CLIP and Aes are external to that loss; CMs winning on these metrics is a substantive, non-circular finding. The ODE-solving half is circular under the rubric's self-definitional pattern: Equation (5) trains Direct CMs to minimize the expected distance d(fθ(xtn,tn), fsolver(xtn,tn,0)) over all timesteps, and Equation (6) defines the headline 'ODE solving error' E as the same distance at t=T. A lower E for Direct CMs is therefore substantially guaranteed by the objective, not an independent measurement of how accurately the empirical PF ODE is solved. No verification that fsolver is close to f^s (the exact solution of Equation 3) or to f^* is supplied, so the counterexample as stated ('better ODE solving does not imply better samples') does not separate ODE fidelity from solver-matching. No load-bearing self-citation was found: [22] and [50] include overlapping authors but support background and evaluation-caveat points, not the central derivation. Because one premise of the main dissociation reduces by construction while the sample-quality comparison retains independent content, the appropriate score is 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central empirical comparison inherits several unverified premises: that the numerical solver is a valid stand-in for the exact ODE solution, that 250 LoRA steps are enough for both losses to converge, and that the chosen image metrics are the right arbiter. The only fitted quantities are standard hyperparameters shared by both methods; no parameter is tuned to force the observed dissociation. The main circularity is evaluative rather than parametric: the metric E used to define better ODE solving is a term in the Direct CM training loss.

free parameters (4)
  • Loss weighting λ(t_n) = 1
    Constant weighting used in both LCD (Equation 4) and LDirect (Equation 5), listed in Table 2. It is chosen by hand and identical across methods, so it does not drive the comparison.
  • Boundary condition schedule (σ_data, τ) = 0.5, 10
    Defines cskip(t) and cout(t) following Luo et al. (Table 2, Appendix A.1). These hand-chosen hyperparameters shape the student parameterization; Direct CMs use the same schedule for control.
  • LoRA rank r and alpha α = 64, 64
    Sets student capacity in both methods (Table 2). Chosen from prior LCM practice, not swept in this paper; any effect on the comparison is held constant.
  • Number of training steps = 250
    Both models trained for 250 steps, with the authors noting curves stabilized around this point (Appendix A.1). This stopping choice could matter: Direct CMs may need more steps to convert lower ODE error into better images, which would weaken the conclusion.
assumptions (5)
  • standard math Under standard regularity conditions, the probability flow ODE has unique trajectories and defines the deterministic mapping f*.
    Taken from Song et al. [48] in Section 2; not re-derived. It justifies treating any point on a trajectory as sufficient to determine the whole trajectory.
  • domain assumption The empirical PF ODE with learned score s, including classifier-free guidance, is the object that consistency models should solve.
    Central to the paper's framing (Equations 3-5). If s is far from the true score, matching fsolver may not mean matching the true generative ODE; the paper itself raises this in Section 5(ii).
  • domain assumption A numerical solver with N intervals is a valid proxy for the exact solution of the empirical PF ODE.
    E (Equation 6) compares only against fsolver, not the exact solution; no convergence or error estimate of the solver is reported. This is the load-bearing premise behind Direct CMs solve the ODE better.
  • domain assumption Both models are trained sufficiently close to their optimal losses for a fair comparison.
    Both are trained for only 250 LoRA steps (Appendix A.1) with stabilization noted but no convergence evidence. Undertrained Direct CMs could explain worse samples without contradicting the loss comparison.
  • domain assumption FID, FD-DINO, CLIP, and aesthetic score are accepted proxies for sample quality.
    The worse sample quality conclusion depends on these metrics; at four steps, FD-DINO and CLIP favor Direct CMs, so the conclusion is metric-dependent (Appendix A.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inconsistencies In Consistency Models: Better ODE Solving Does Not Imply Better Samples." pith.science (2026). https://pith.science/paper/4DFWFHTR

@misc{pith2026241108954,
  author       = {Pith},
  title        = {Pith review of: Inconsistencies In Consistency Models: Better ODE Solving Does Not Imply Better Samples},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4DFWFHTR}},
  note         = {Machine review of arXiv:2411.08954}
}
read the original abstract

Although diffusion models can generate remarkably high-quality samples, they are intrinsically bottlenecked by their expensive iterative sampling procedure. Consistency models (CMs) have recently emerged as a promising diffusion model distillation method, reducing the cost of sampling by generating high-fidelity samples in just a few iterations. Consistency model distillation aims to solve the probability flow ordinary differential equation (ODE) defined by an existing diffusion model. CMs are not directly trained to minimize error against an ODE solver, rather they use a more computationally tractable objective. As a way to study how effectively CMs solve the probability flow ODE, and the effect that any induced error has on the quality of generated samples, we introduce Direct CMs, which \textit{directly} minimize this error. Intriguingly, we find that Direct CMs reduce the ODE solving error compared to CMs but also result in significantly worse sample quality, calling into question why exactly CMs work well in the first place. Full code is available at: https://github.com/layer6ai-labs/direct-cms.

Figures

Figures reproduced from arXiv: 2411.08954 by the authors.

Figure 1
Figure 1. CMs (left) are weakly supervised ODE solvers, only learning to map points along a [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Samples generated by both CMs and Direct CMs. The samples produced by CMs are [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Effect of the teacher’s number of discretization intervals [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Effect of the teacher’s guidance scale ω. We use N = 50 here for faster experimentation. In all cases, we observe that Direct CMs are better at solving the PF ODE, but CMs produce higher quality images. rely on auxiliary losses to supplement ODE solving such as adversa…
Figure 5
Figure 5. Figure 5: Additional images generated by both CMs and Direct CMs, further highlighting the sample [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 26 canonical work pages

  1. [1]

    Tract: Denoising diffusion models with transitive closure time-distillation

    David Berthelot, Arnaud Autef, Jierui Lin, Dian Ang Yap, Shuangfei Zhai, Siyuan Hu, Daniel Zheng, Walter Talbott, and Eric Gu. Tract: Denoising diffusion models with transitive closure time-distillation. arXiv:2303.04248, 2023

    arXiv 2023

  2. [2]

    Stable video diffusion: Scaling latent video diffusion models to large datasets

    Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram V oleti, Adam Letts, Varun Jampani, and Robin Rombach. Stable video diffusion: Scaling latent video diffusion models to large datasets. arXiv:2311.15127, 2023

    arXiv 2023

  3. [3]

    Flash diffusion: Accelerating any conditional diffusion model for few steps image generation

    Clement Chadebec, Onur Tasar, Eyal Benaroche, and Benjamin Aubin. Flash diffusion: Accelerating any conditional diffusion model for few steps image generation. arXiv:2406.02347, 2024

    arXiv 2024

  4. [4]

    Sampling is as easy as learning the score: Theory for diffusion models with minimal data assumptions

    Sitan Chen, Sinho Chewi, Jerry Li, Yuanzhi Li, Adil Salim, and Anru Zhang. Sampling is as easy as learning the score: Theory for diffusion models with minimal data assumptions. In The Eleventh International Conference on Learning Representations , 2023

    work page 2023

  5. [5]

    Emu: Enhancing image generation models using photogenic needles in a haystack

    Xiaoliang Dai, Ji Hou, Chih-Yao Ma, Sam Tsai, Jialiang Wang, Rui Wang, Peizhao Zhang, Simon Vandenhende, Xiaofang Wang, Abhimanyu Dubey, Matthew Yu, Abhishek Kadian, Filip Radenovic, Dhruv Mahajan, Kunpeng Li, Yue Zhao, Vladan Petrovic, Mitesh Kumar Singh, Simran Motwani, Yi Wen, Yiwen Song, Roshan Sumbaly, Vignesh Ramanathan, Zijian He, Peter Vajda, and ...

    arXiv 2023

  6. [6]

    8-bit optimizers via block-wise quantiza- tion

    Tim Dettmers, Mike Lewis, Sam Shleifer, and Luke Zettlemoyer. 8-bit optimizers via block-wise quantiza- tion. In International Conference on Learning Representations , 2022

    work page 2022

  7. [7]

    Consistency models made easy

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

    arXiv 2024

  8. [8]

    Griffiths and Desmond J

    David F. Griffiths and Desmond J. Higham. Numerical Methods for Ordinary Differential Equations . Springer, 2010

    work page 2010

Show all 57 references
  1. [9]

    Boot: Data-free distillation of denoising diffusion models with bootstrapping

    Jiatao Gu, Shuangfei Zhai, Yizhe Zhang, Lingjie Liu, and Joshua M Susskind. Boot: Data-free distillation of denoising diffusion models with bootstrapping. In ICML 2023 Workshop on Structured Probabilistic Inference & Generative Modeling, 2023

    2023

  2. [10]

    Multistep consistency models

    Jonathan Heek, Emiel Hoogeboom, and Tim Salimans. Multistep consistency models. arXiv:2403.06807, 2024

    2024 arXiv

  3. [11]

    CLIPScore: A Reference-free Evaluation Metric for Image Captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. CLIPScore: A Reference-free Evaluation Metric for Image Captioning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021. doi: 10.18653/v1/2021.emnlp-main.595

    2021 doi

  4. [12]

    GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium. In Advances in Neural Information Processing Systems, volume 30, 2017

    2017

  5. [13]

    Classifier-free diffusion guidance

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv:2207.12598, 2022

    2022 arXiv

  6. [14]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In Advances in Neural Information Processing Systems, volume 33, 2020

    2020

  7. [15]

    Video diffusion models

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video diffusion models. In Advances in Neural Information Processing Systems , volume 35, 2022

    2022

  8. [16]

    LoRA: Low-Rank Adaptation of Large Language Models

    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. In International Conference on Learning Representations, 2022

    2022

  9. [17]

    Make-an-audio: Text-to-audio generation with prompt-enhanced diffusion models

    Rongjie Huang, Jiawei Huang, Dongchao Yang, Yi Ren, Luping Liu, Mingze Li, Zhenhui Ye, Jinglin Liu, Xiang Yin, and Zhou Zhao. Make-an-audio: Text-to-audio generation with prompt-enhanced diffusion models. In Proceedings of the 40th International Conference on Machine Learning , 2023

    2023

  10. [18]

    Consistency trajectory models: Learning probability flow ODE trajectory of diffusion

    Dongjun Kim, Chieh-Hsin Lai, Wei-Hsiang Liao, Naoki Murata, Yuhta Takida, Toshimitsu Uesaka, Yutong He, Yuki Mitsufuji, and Stefano Ermon. Consistency trajectory models: Learning probability flow ODE trajectory of diffusion. In The Twelfth International Conference on Learning ...

    2024

  11. [19]

    PaGoDA: Progressive Growing of a One-Step Generator from a Low-Resolution Diffusion Teacher

    Dongjun Kim, Chieh-Hsin Lai, Wei-Hsiang Liao, Yuhta Takida, Naoki Murata, Toshimitsu Uesaka, Yuki Mitsufuji, and Stefano Ermon. PaGoDA: Progressive Growing of a One-Step Generator from a Low-Resolution Diffusion Teacher. arXiv:2405.14822, 2024. 7

    2024 arXiv

  12. [20]

    Diffwave: A versatile diffusion model for audio synthesis

    Zhifeng Kong, Wei Ping, Jiaji Huang, Kexin Zhao, and Bryan Catanzaro. Diffwave: A versatile diffusion model for audio synthesis. In International Conference on Learning Representations , 2021

    2021

  13. [21]

    Bidirectional consistency models

    Liangchen Li and Jiajun He. Bidirectional consistency models. arXiv:2403.18035, 2024

    2024 arXiv

  14. [22]

    Deep generative models through the lens of the manifold hypothesis: A survey and new connections

    Gabriel Loaiza-Ganem, Brendan Leigh Ross, Rasa Hosseinzadeh, Anthony L Caterini, and Jesse C Cresswell. Deep generative models through the lens of the manifold hypothesis: A survey and new connections. Transactions on Machine Learning Research, 2024

    2024

  15. [23]

    Mathematical analysis of singularities in the diffusion model under the submanifold assumption

    Yubin Lu, Zhongjian Wang, and Guillaume Bal. Mathematical analysis of singularities in the diffusion model under the submanifold assumption. arXiv:2301.07882, 2023

    2023 arXiv

  16. [24]

    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:2101.02388, 2021

    2021 arXiv

  17. [25]

    Latent consistency models: Synthesizing high-resolution images with few-step inference

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high-resolution images with few-step inference. arXiv:2310.04378, 2023

    2023 arXiv

  18. [26]

    LCM-LoRA: A universal stable-diffusion acceleration module

    Simian Luo, Yiqin Tan, Suraj Patil, Daniel Gu, Patrick von Platen, Apolinário Passos, Longbo Huang, Jian Li, and Hang Zhao. LCM-LoRA: A universal stable-diffusion acceleration module. arXiv:2311.05556, 2023

    2023 arXiv

  19. [27]

    On distillation of guided diffusion models

    Chenlin Meng, Robin Rombach, Ruiqi Gao, Diederik Kingma, Stefano Ermon, Jonathan Ho, and Tim Salimans. On distillation of guided diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14297–14306, 2023

    2023

  20. [28]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel HAZIZA, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rab...

    2024

  21. [29]

    Score-based generative models detect manifolds

    Jakiw Pidstrigach. Score-based generative models detect manifolds. In Advances in Neural Information Processing Systems, volume 35, 2022

    2022

  22. [30]

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

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis. In The Twelfth International Conference on Learning Representations , 2024

    2024

  23. [31]

    Barron, and Ben Mildenhall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Mildenhall. Dreamfusion: Text-to-3d using 2d diffusion. In The Eleventh International Conference on Learning Representations , 2023

    2023

  24. [32]

    Self-attention Does Not Need O(n2) Memory

    Markus N Rabe and Charles Staats. Self-attention Does Not Need O(n2) Memory. arXiv:2112.05682, 2021

    2021 arXiv

  25. [33]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. In Proceedings of th...

    2021

  26. [34]

    Hierarchical text-conditional image generation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv:2204.06125, 2022

    2022 arXiv

  27. [35]

    Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis

    Yuxi Ren, Xin Xia, Yanzuo Lu, Jiacheng Zhang, Jie Wu, Pan Xie, Xing Wang, and Xuefeng Xiao. Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis. arXiv:2404.13686, 2024

    2024 arXiv

  28. [36]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022

    2022

  29. [37]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part ...

    2015

  30. [38]

    André Ronveaux and F. M. Arscott. Heun’s Differential Equations. Oxford University Press, 1995. 8

    1995

  31. [39]

    MM-Diffusion: Learning Multi-Modal Diffusion Models for Joint Audio and Video Generation

    Ludan Ruan, Yiyang Ma, Huan Yang, Huiguo He, Bei Liu, Jianlong Fu, Nicholas Jing Yuan, Qin Jin, and Baining Guo. MM-Diffusion: Learning Multi-Modal Diffusion Models for Joint Audio and Video Generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern R...

    2023

  32. [40]

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

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, Jonathan Ho, David J Fleet, and Mohammad Norouzi. Photorealistic text-to-image diffusion models with deep language un...

    2022

  33. [41]

    Progressive distillation for fast sampling of diffusion models

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models. In International Conference on Learning Representations , 2022

    2022

  34. [42]

    Adversarial diffusion distillation

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distillation. arXiv:2311.17042, 2023

    2023 arXiv

  35. [43]

    LAION-5B: An open large-scale dataset for training next generation image-text models

    Christoph Schuhmann, Romain Beaumont, Richard Vencu, Cade Gordon, Ross Wightman, Mehdi Cherti, Theo Coombes, Aarush Katta, Clayton Mullis, Mitchell Wortsman, Patrick Schramowski, Srivatsa Kun- durthy, Katherine Crowson, Ludwig Schmidt, Robert Kaczmarczyk, and Jenia Jitsev. LAI...

    2022

  36. [44]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Proceedings of the 32nd International Conference on Machine Learning, volume 37, 2015

    2015

  37. [45]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. In International Conference on Learning Representations, 2021

    2021

  38. [46]

    Improved techniques for training consistency models

    Yang Song and Prafulla Dhariwal. Improved techniques for training consistency models. In The Twelfth International Conference on Learning Representations , 2024

    2024

  39. [47]

    Improved techniques for training score-based generative models

    Yang Song and Stefano Ermon. Improved techniques for training score-based generative models. In Advances in Neural Information Processing Systems , volume 33, 2020

    2020

  40. [48]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021

    2021

  41. [49]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. In Proceedings of the 40th International Conference on Machine Learning , volume 202, 2023

    2023

  42. [50]

    Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models

    George Stein, Jesse C Cresswell, Rasa Hosseinzadeh, Yi Sui, Brendan Ross, Valentin Villecroze, Zhaoyan Liu, Anthony L Caterini, J Eric T Taylor, and Gabriel Loaiza-Ganem. Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models. In A...

    2023

  43. [51]

    Phased consistency model

    Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman, Dazhong Shen, Peng Gao, Michael Lingelbach, Keqiang Sun, Weikang Bian, Guanglu Song, Yu Liu, Hongsheng Li, and Xiaogang Wang. Phased consistency model. arXiv:2405.18407, 2024

    2024 arXiv

  44. [52]

    VideoComposer: Compositional Video Synthesis with Motion Controllability

    Xiang Wang, Hangjie Yuan, Shiwei Zhang, Dayou Chen, Jiuniu Wang, Yingya Zhang, Yujun Shen, Deli Zhao, and Jingren Zhou. VideoComposer: Compositional Video Synthesis with Motion Controllability. In Advances in Neural Information Processing Systems , volume 36, 2023

    2023

  45. [53]

    Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation

    Jay Zhangjie Wu, Yixiao Ge, Xintao Wang, Stan Weixian Lei, Yuchao Gu, Yufei Shi, Wynne Hsu, Ying Shan, Xiaohu Qie, and Mike Zheng Shou. Tune-a-video: One-shot tuning of image diffusion models for text-to-video generation. In Proceedings of the IEEE/CVF International Conference...

    2023

  46. [54]

    MLCM: Multistep Consistency Distillation of Latent Diffusion Model

    Qingsong Xie, Zhenyi Liao, Chen chen, Zhijie Deng, Shixiang Tang, and Haonan Lu. MLCM: Multistep Consistency Distillation of Latent Diffusion Model. arXiv:2406.05768, 2024

    2024 arXiv

  47. [55]

    Freeman, and Taesung Park

    Tianwei Yin, Michaël Gharbi, Richard Zhang, Eli Shechtman, Frédo Durand, William T. Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024. 9

    2024

  48. [56]

    Fast sampling of diffusion models via operator learning

    Hongkai Zheng, Weili Nie, Arash Vahdat, Kamyar Azizzadenesheli, and Anima Anandkumar. Fast sampling of diffusion models via operator learning. In Proceedings of the 40th International Conference on Machine Learning, volume 202, 2023

    2023

  49. [57]

    Trajectory consistency distillation

    Jianbin Zheng, Minghui Hu, Zhongyi Fan, Chaoyue Wang, Changxing Ding, Dacheng Tao, and Tat-Jen Cham. Trajectory consistency distillation. arXiv:2402.19159, 2024. 10 A Appendix A.1 Hyperparameters We provide a list of default hyperparameter values in Table 2. We only train a sm...

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.