Pith. sign in

REVIEW 3 major objections 5 minor 87 references

One-Way Ticket:Time-Independent Unified Encoder for Distilling Text-to-Image Diffusion Models

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A distilled text-to-image UNet can drop the sequential denoising loop: run the encoder once and let K parallel decoders finish the image, preserving quality and diversity at near one-step cost.

desk verdict One-pass shared encoder with parallel decoder steps is a real, diversity-improving contribution, but the paper's ODE-trajectory claim is contradicted by its own equations and needs correcting. read the letter →

arxiv 2505.21960 v1 pith:C47Q36DU submitted 2025-05-28 cs.CV

Senmao Li , Lei Wang , Kai Wang , Tao Liu , Jiehang Xie , Joost van de Weijer , Fahad Shahbaz Khan , Shiqi Yang
show 2 more authors
Yaxing Wang Jian Yang
This is my paper
classification cs.CV
keywords text-to-imagediffusiondistillationloop-freesamplingparalleldenoisingUNetencodersharingvariationalscoreKLdivergenceregularizationdiverseimagegeneration
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

This paper sets out to show that the redundant encoder of a text-to-image diffusion UNet can be run once and then shared by several decoder passes at different time steps, turning sequential denoising into a single parallel pass. The proposed student architecture, TiUE, computes a shared encoder feature from the initial noise and then evaluates the decoder at $K$ time steps in parallel, so inference time stays comparable to one-step generators even though the decoder visits multiple time steps. On the 30K-prompt COCO 2014 benchmark the paper reports FID 13.09 and Recall 0.622 for TiUE, against FID 15.98 and Recall 0.457 for SwiftBrushv2 and 19.51 and 0.458 for SD-Turbo at one step. A KL-divergence term keeps the predicted noise close to a standard Gaussian, which they report improves both quality and diversity while keeping training free of real or teacher-synthesized images. If the central claim holds, the practical consequence is fast, diverse text-to-image generation without the usual quality-diversity trade-off.

What carries the argument

The central object is the Time-independent Unified Encoder (TiUE), a student UNet in which the encoder runs once and the decoder is shared across $K$ time steps. The identity that makes the design work is the factorization of the DDIM update into independent terms: because the decoder's input is the fixed feature $f$ rather than the latent $z_t$, the predicted noises $\epsilon_1, \ldots, \epsilon_K$ can be computed in parallel, and the final latent is a weighted sum of the input noise and those predictions. The skip-connection and mid-block features produced by the single encoder pass are reused by every decoder step, which is what removes the sequential loop. A KL divergence toward $\mathcal{N}(0, I)$ regulates the student's output distribution, and variational score distillation (VSD) with an alternating SD-LoRA update supplies the training signal.

What would settle it

Measure FID after restoring the standard teacher input at each step: if giving the decoder the current latent $z_t$ markedly improves quality at equal compute, the fixed-feature premise is refuted. Alternatively, swap the shared feature $f$ for one computed from a different latent $z_{t'}$ and check whether the predicted $\epsilon_t$ and the final image change appreciably, which would show the encoder feature is not actually time-independent.

Watch

Extended reading notes

Core claim

The paper's central claim is that the time-dependent encoder of a UNet is unnecessary at every step once the student is distilled. The student generator $\epsilon^{SG}_{\theta}$ takes a Gaussian noise $\epsilon$, a text prompt $y$, and a key time $K$, and runs its encoder exactly once, producing a shared feature map $f = \epsilon^{SG\text{-}EN}_{\theta}(\epsilon, K, y)$. Each decoder step then computes the predicted noise $\epsilon_t = \epsilon^{SG\text{-}DE}_{\theta}(f, t, y)$ for $t = K, \ldots, 1$ with no dependence on the evolving latent $z_t$, so all decoder passes can be evaluated in parallel. Substituting this into the DDIM update yields the final latent as a fixed linear combination of the input noise and the parallel predictions, $z_0 = S\epsilon + \sum_{t=1}^{K} E_t \epsilon_t$, with coefficients fixed by the noise schedule. The paper argues this design stays on the teacher's DDIM trajectory while letting the decoder visit multiple time steps, which is why outputs remain semantically rich and diverse; the KL-divergence term pushes predicted noise toward $\mathcal{N}(0, I)$ and is credited with improving realism and diversity without requiring image data.

Load-bearing premise

The whole scheme rests on the premise that a single encoder feature, computed once from the initial noise, carries enough information to predict the noise at every later decoder step even though the decoder never sees the current partially denoised latent.

Editorial extensions

If this is right

  • One-pass inference cuts the UNet encoder cost to a single forward pass for $K$ decoder steps, so time-to-image approaches one-step models while the decoder still sees multiple time steps.
  • The KL-regularization term improves FID, precision, density, and coverage in the paper's ablation, indicating that preserving the Gaussian noise prior helps diversity as well as fidelity.
  • Because training uses only text captions and no real or teacher-synthesized images, the distillation recipe can be applied where paired image-text data is scarce.
  • A model trained with $K=4$ still generates coherent images when sampled with other numbers of steps, suggesting the student retains useful trajectory knowledge from the pretrained model.
  • With encoder sharing, the memory footprint grows modestly while latency stays close to one-step baselines, which favors deployment over sequential multi-step distillation.

Reading between the lines

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

  • A natural testable extension is to let the shared feature $f$ depend weakly on the time step $t$ through a lightweight adapter, restoring some time-dependence at small extra cost and potentially closing any residual gap to true ODE following.
  • If the one-pass premise holds for latent UNets, the same idea could transfer to transformer-based diffusion backbones, where an analogous single context pass might parallelize token-level denoising.
  • The linear decomposition $z_0 = S\epsilon + \sum E_t \epsilon_t$ invites a direct ablation: dropping individual $\epsilon_t$ terms reveals which decoder time steps carry the semantic information that produces diversity.
  • The KL term may act as a mode-cover regularizer; combining it with adversarial or consistency losses could produce one-pass models with sharper texture without sacrificing the diversity reported here.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes TiUE, a distillation scheme for text-to-image latent diffusion models in which the student UNet encoder is evaluated once on the initial noise, while the student decoder is applied in parallel at K time steps using shared weights and the same encoder feature. Training uses variational score distillation (VSD) with an SD-LoRA teacher, plus a KL divergence term intended to keep the student's noise predictions close to a standard Gaussian. The authors report one-pass inference with quality and diversity improvements over LCM, SD-Turbo, SwiftBrush, and SwiftBrushv2 on COCO 2014/2017, AFHQ, CelebA-HQ, DrawBench, and PartiPrompts, along with a user study and various ablations.

Significance. If the claims hold, the paper makes a practical contribution: a loop-free, one-encoder-pass architecture that parallelizes decoder evaluations and improves diversity relative to existing one-step distilled models. The empirical evaluation is broad and includes diversity-oriented metrics (Recall, Density, Coverage) that are often missing from distillation papers, and the reported diversity gains over SD-Turbo and SwiftBrushv2 are concrete. The image-free training setup, using only captions from JourneyDB, is also a practical strength. However, the theoretical framing is currently not sound: the paper's claim of adhering to the teacher's ODE trajectory is contradicted by its own inference formula, and the KL regularizer as written has a sign error. These issues are load-bearing for the paper's central narrative and need to be corrected before the empirical results can be fully credited.

major comments (3)
  1. [§3.2, Eq. (5) and Appendix A.5, Eqs. (7)–(11)] The sentence "our method TiUE also adheres to the original ODE trajectory" is not supported by the inference formula. In the teacher's DDIM update (Eq. (1)) the denoiser is evaluated at the current latent z_t, which encodes accumulated information from previous steps. In the student's inference, each ε_t = ε^SG-DE_θ(f, t, y) is evaluated from the same encoder feature f computed once from the initial noise, with no dependence on the evolving latent z_t. Consequently, Eq. (11) is a one-shot linear combination of four noise estimates evaluated at the same initial noise, not a discretization of the teacher's reverse ODE. This is a load-bearing point because the paper motivates the architecture as trajectory-preserving distillation. The method may still train a useful one-pass generator via VSD, but the stated theoretical justification must be removed or replaced with a direct argument for why f is a sufficient conditioning signal. In particular, because the decoder never sees z_t, the K predictions cannot correct one another, so the parallel decoder cannot be presented as following the teacher trajectory in the claimed sense.
  2. [§3.2, Eq. (6)] The definition D_KL(p||q) = -∫ p(z) log(p(z)/q(z)) dz has the wrong sign. The standard KL divergence is ∫ p(z) log(p(z)/q(z)) dz and is nonnegative, while the definition in the paper makes L_KL the negative of the standard KL. Minimizing that loss would drive the student's noise predictions away from N(0,I), which contradicts the stated purpose of regularizing the noise prediction toward a Gaussian (Sec. 4.2, Fig. 8). Additionally, Eq. (6) is not well-defined as written because ϵ^SG_θ(ϵ, K, t, y) is a single deterministic network output rather than a distribution; the authors need to specify the distributional estimator they use, for example an empirical distribution over a batch. This is central to the claimed contribution of KL-based regularization.
  3. [§3.2, Fig. 1 and Fig. 2] The load-bearing assumption that encoder features computed once at time K are sufficient conditioning for all K decoder passes is supported only by qualitative observations. Figure 1 shows feature similarity across time steps in the teacher model, and Figure 2 shows decoder features over time in a distilled model, but neither demonstrates that a single encoder feature from pure noise can replace the time-evolving latent for every step t. Because Eq. (11) has no feedback from intermediate latents, the success of the method depends entirely on this hypothesis. The paper should provide a direct quantitative test, such as ablating K with the same encoder (e.g., K=1, 2, 3, 4) or comparing the one-pass encoder against a variant that recomputes the encoder at each step. Without such evidence, the architectural claim is under-supported.
minor comments (5)
  1. [§4.1] The text "Our approach outperforms LCM and SwiftBrush across all evaluation metrics, except for Precision" appears to contain a naming error: there is no 4-step SwiftBrush row in Table 1, and the sentence should presumably read "LCM and SD-Turbo."
  2. [§4.2 and Fig. 8 (right)] The figure reporting the predicted noise distribution lacks axis labels and a description of how the fitted curve was obtained; please specify what quantity is plotted and over which samples the distribution is estimated.
  3. [Table 2] For DrawBench and PartiPrompts, SD2.1 samples are used as the ground truth, which makes the SD2.1 row trivially have FID=0 and Density/Coverage=1; the text should state more prominently that these numbers measure agreement with SD2.1 rather than absolute diversity.
  4. [Eq. (6)] The notation T1 in ϵ^SG_θ(ϵ, T1, t, y) is unexplained and inconsistent with the earlier use of K as the encoder time step; presumably it should be K, and this should be corrected.
  5. [Introduction and Appendix A.5] The paper repeatedly claims "the first Time-independent Unified Encoder" and calls the method "one-step inference"; given the difficulty of verifying novelty and the fact that inference uses K=4 decoder passes, these phrases should be softened or more carefully qualified.

Circularity Check

1 steps flagged · score 2.0 of 10

No fitted-input circularity; one definitional ODE-trajectory claim overstates the one-pass update, and the Faster Diffusion self-citation is not load-bearing.

  1. self definitional [Section 3.2, Eq. (5) and Appendix A.5, Eqs. (7)-(11)]
    "The Eq. (1) can be rewritten as: zt−1= q αt−1 αt zt +√αt−1 hq 1 αt−1 −1− q 1 αt −1 i ·ϵSGθ (ϵ, K, t, y). (5) By this means, our method TiUE also adheres to the original ODE trajectory [42, 67]."

    Eq. (5) is presented as a rewrite of the teacher DDIM update Eq. (1), but the student denoiser is evaluated at the fixed initial noise ϵ instead of the current latent zt. The 'adherence to the original ODE trajectory' is therefore attached by definition to a DDIM-looking coefficient schedule with a different network argument, not derived from the teacher ODE. The actual inference (Eqs. 7-11) is a one-shot linear combination z0 = Sϵ + E4ϵ4 + E3ϵ3 + E2ϵ2 + E1ϵ1 of noise estimates that never see intermediate latents, so it is a parallel one-pass map rather than a discretization of the teacher's reverse ODE. This is a definitional labeling issue, not a fitted-input reduction, and the experimental results are externally benchmarked.

full rationale

The paper's empirical contribution is self-contained: TiUE is trained with VSD and KL losses and evaluated against external one-step baselines on COCO, AFHQ, CelebA-HQ, DrawBench, and PartiPrompts, so no fitted parameter is relabeled as a prediction. The motivational observation about encoder feature sharing is partly self-cited (Faster Diffusion [34] shares the first author), but the paper re-derives the correlation in Fig. 1 and Fig. 2 and cites independent feature-analysis works (PnP, DIFT), so the self-citation is not load-bearing in the circularity sense. The one definitional overreach is the claim that Eq. (5) 'adheres to the original ODE trajectory': the update is obtained from Eq. (1) by replacing the teacher's state input zt with the fixed initial noise ϵ, and the resulting parallel combination (Eqs. 7-11) is not a discretization of the teacher's reverse ODE. That is an over-claimed label attached by construction rather than a derived equivalence; it does not infect the experimental contribution, which stands or falls on external benchmark results. Hence the low score.

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

The central result rests on a small number of design choices. K=4 is a hand-set architectural parameter, not swept. The encoder-sufficiency assumption is the load-bearing leap: all K decoder noise predictions are functions of the single encoder output f and do not depend on the evolving latent. The KL regularizer is an empirically motivated addition, but its estimator and weight are unspecified. No new physical or latent entities are introduced.

free parameters (2)
  • K (number of decoder time steps) = 4
    Chosen by analogy with 4-step LCM/SD-Turbo schedules and not swept; the paper notes 'we still achieve better results with K=4' but provides no K-ablation table.
  • KL divergence loss weight = not reported
    Eq. (6) defines L_KL but no coefficient is given for the sum with L_vsd; without it the method cannot be reproduced as stated.
assumptions (4)
  • standard math The DDIM update (Eq. 1) is a valid discretization of the reverse ODE.
    The student's recursive latent updates in Appendix A.5 rely on the DDIM coefficients from Song et al. [67].
  • ad hoc to paper Encoder features computed once at time K are sufficient conditioning for all K decoder passes.
    Sec. 3.2 and Appendix A.5 define eps_t = eps^SG-DE(f, t, y) with no dependence on the current latent zt; this is the load-bearing assumption for loop-free parallel decoding.
  • ad hoc to paper D_KL of the student's noise output against N(0,I) is a computable and useful regularizer.
    Eq. (6) introduces L_KL, but the paper never states how the empirical KL is estimated from a batch nor what weight to apply.
  • domain assumption VSD with alternating LoRA updates provides correct distillation gradients.
    The training procedure follows SwiftBrush [53] and ProlificDreamer [78]; convergence to the teacher distribution is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of One-Way Ticket:Time-Independent Unified Encoder for Distilling Text-to-Image Diffusion Models." pith.science (2026). https://pith.science/paper/C47Q36DU

@misc{pith2026250521960,
  author       = {Pith},
  title        = {Pith review of: One-Way Ticket:Time-Independent Unified Encoder for Distilling Text-to-Image Diffusion Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C47Q36DU}},
  note         = {Machine review of arXiv:2505.21960}
}
read the original abstract

Text-to-Image (T2I) diffusion models have made remarkable advancements in generative modeling; however, they face a trade-off between inference speed and image quality, posing challenges for efficient deployment. Existing distilled T2I models can generate high-fidelity images with fewer sampling steps, but often struggle with diversity and quality, especially in one-step models. From our analysis, we observe redundant computations in the UNet encoders. Our findings suggest that, for T2I diffusion models, decoders are more adept at capturing richer and more explicit semantic information, while encoders can be effectively shared across decoders from diverse time steps. Based on these observations, we introduce the first Time-independent Unified Encoder TiUE for the student model UNet architecture, which is a loop-free image generation approach for distilling T2I diffusion models. Using a one-pass scheme, TiUE shares encoder features across multiple decoder time steps, enabling parallel sampling and significantly reducing inference time complexity. In addition, we incorporate a KL divergence term to regularize noise prediction, which enhances the perceptual realism and diversity of the generated images. Experimental results demonstrate that TiUE outperforms state-of-the-art methods, including LCM, SD-Turbo, and SwiftBrushv2, producing more diverse and realistic results while maintaining the computational efficiency.

Figures

Figures reproduced from arXiv: 2505.21960 by the authors.

Figure 1
Figure 1. The correlation between image generation quality (Fig.a-b) and encoder feature similarity (Fig.c). Above a certain threshold of [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The generated images and the corresponding UNet de [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Generated high-fidelity 5122 images by the one-step model distilled with our method TiUE. Compared with the baselines, our method produces higher image quality and more diverse results. Here the diversity means the degree of variation of different output images with same semantic information when given the same prompt and different seed values. SwiftBrushv2 [7], which is initialized with SD￾Turbo [64], achieves T2I … view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Previous methods use the time-dependent UNet encoder [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Loop-free distillation based on our Time-independent Unified Encoder ( [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison to state-of-the-art few-step distillation methods. [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 8
Figure 8. Figure 8: User Study (left). Comparison of the predicted noise [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 10
Figure 10. Figure 10: Results of different time-steps with K=4 trained student model. The well-trained student model with time-steps K=4 (the second column) serves as a good starting point for students set at other time-steps (e.g., 2, 8, 16, 32, and 50). in [PITH_FULL_IMAGE:figures/full_…
Figure 9
Figure 9. Figure 9: Interpolation between text prompts. For example, in [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

87 extracted references · 59 canonical work pages

  1. [1]

    Token merging for fast sta- ble diffusion

    Daniel Bolya and Judy Hoffman. Token merging for fast sta- ble diffusion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 4598– 4602, 2023. 1, 3

    2023

  2. [2]

    Colorpeel: Color prompt learning with diffusion models via color and shape disentanglement

    Muhammad Atif Butt, Kai Wang, Javier Vazquez-Corral, and Joost van de Weijer. Colorpeel: Color prompt learning with diffusion models via color and shape disentanglement. In ECCV, 2024. 1

    2024

  3. [3]

    Q-dit: Ac- curate post-training quantization for diffusion transformers

    Lei Chen, Yuan Meng, Chen Tang, Xinzhu Ma, Jingyan Jiang, Xin Wang, Zhi Wang, and Wenwu Zhu. Q-dit: Ac- curate post-training quantization for diffusion transformers. CVPR, 2025. 3

    2025

  4. [4]

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

    Pengtao Chen, Mingzhu Shen, Peng Ye, Jianjian Cao, Chongjun Tu, Christos-Savvas Bouganis, Yiren Zhao, and Tao Chen. Delta-dit: A training-free acceleration method tailored for diffusion transformers.arXiv preprint arXiv:2406.01125, 2024. 3

    arXiv 2024

  5. [5]

    Asyncdiff: Parallelizing diffusion mod- els by asynchronous denoising.NeurIPS, 2024

    Zigeng Chen, Xinyin Ma, Gongfan Fang, Zhenxiong Tan, and Xinchao Wang. Asyncdiff: Parallelizing diffusion mod- els by asynchronous denoising.NeurIPS, 2024. 3

    2024

  6. [6]

    Stargan v2: Diverse image synthesis for multiple domains

    Yunjey Choi, Youngjung Uh, Jaejun Yoo, and Jung-Woo Ha. Stargan v2: Diverse image synthesis for multiple domains. InProceedings of the IEEE/CVF conference on computer vi- sion and pattern recognition, pages 8188–8197, 2020. 5, 7, 1

    2020

  7. [7]

    Swiftbrush v2: Make your one-step diffusion model better than its teacher.ECCV,

    Trung Dao, Thuan Hoang Nguyen, Thanh Le, Duc Vu, Khoi Nguyen, Cuong Pham, and Anh Tran. Swiftbrush v2: Make your one-step diffusion model better than its teacher.ECCV,

  8. [8]

    DeepFloyd IF.https://www.deepfloyd

    DeepFloyd. DeepFloyd IF.https://www.deepfloyd. ai/deepfloyd-if, 2023. 1

    work page 2023

Show all 87 references
  1. [9]

    Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.Advances in neural informa- tion processing systems, 34:8780–8794, 2021. 1

    2021

  2. [10]

    Structural pruning for diffusion models.Advances in neural informa- tion processing systems, 36, 2024

    Gongfan Fang, Xinyin Ma, and Xinchao Wang. Structural pruning for diffusion models.Advances in neural informa- tion processing systems, 36, 2024. 3

    2024

  3. [11]

    One- step diffusion distillation via deep equilibrium models

    Zhengyang Geng, Ashwini Pokle, and J Zico Kolter. One- step diffusion distillation via deep equilibrium models. In Thirty-seventh Conference on Neural Information Process- ing Systems, 2023. 3

    2023

  4. [12]

    Tokenflow: Consistent diffusion features for consistent video editing.ICLR, 2024

    Michal Geyer, Omer Bar-Tal, Shai Bagon, and Tali Dekel. Tokenflow: Consistent diffusion features for consistent video editing.ICLR, 2024. 1

    2024

  5. [13]

    Photoswap: Personalized subject swapping in images, 2023

    Jing Gu, Yilin Wang, Nanxuan Zhao, Tsu-Jui Fu, Wei Xiong, Qing Liu, Zhifei Zhang, He Zhang, Jianming Zhang, Hyun- Joon Jung, and Xin Eric Wang. Photoswap: Personalized subject swapping in images, 2023. 1

    2023

  6. [14]

    Boot: Data-free distillation of denoising dif- fusion models with bootstrapping.International Conference on Machine Learning, 2023

    Jiatao Gu, Shuangfei Zhai, Yizhe Zhang, Lingjie Liu, and Josh Susskind. Boot: Data-free distillation of denoising dif- fusion models with bootstrapping.International Conference on Machine Learning, 2023. 3

    2023

  7. [15]

    Prompt-to-prompt image editing with cross attention control.ICLR, 2023

    Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Prompt-to-prompt image editing with cross attention control.ICLR, 2023. 1

    2023

  8. [16]

    Clipscore: A reference-free evaluation met- ric for image captioning

    Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation met- ric for image captioning. InProceedings of the 2021 Confer- ence on Empirical Methods in Natural Language Processing, pages 7514–7528, 2021. 5, 1

    2021

  9. [17]

    Gans trained by a two time-scale update rule converge to a local nash equilib- rium

    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 equilib- rium. InNeurIPS, pages 6626–6637, 2017. 5, 1

    2017

  10. [18]

    Distilling the Knowledge in a Neural Network.NIPS Deep Learning Workshop, 2014

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the Knowledge in a Neural Network.NIPS Deep Learning Workshop, 2014. 3

    2014

  11. [19]

    Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models.Advances in Neural Information Processing Systems, 33:6840–6851, 2020. 1

    2020

  12. [20]

    Lora: Low-rank adaptation of large language models.ICLR,

    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.ICLR,

  13. [21]

    Token merging for training- free semantic binding in text-to-image synthesis

    Taihang Hu, Linxuan Li, Joost van de Weijer, Hongcheng Gao, Fahad Shahbaz Khan, Jian Yang, Mingming Cheng, Kai Wang, and Yaxing Wang. Token merging for training- free semantic binding in text-to-image synthesis. In NeurIPS, 2024. 1

    2024

  14. [22]

    Harmonica: Harmonizing training and inference for better feature cache in diffusion transformer acceleration.arXiv preprint arXiv:2410.01723, 2024

    Yushi Huang, Zining Wang, Ruihao Gong, Jing Liu, Xin- jie Zhang, Jinyang Guo, Xianglong Liu, and Jun Zhang. Harmonica: Harmonizing training and inference for better feature cache in diffusion transformer acceleration.arXiv preprint arXiv:2410.01723, 2024. 3

    arXiv

  15. [23]

    Contragan: Contrastive learn- ing for conditional image generation.NeurIPS, 2020

    Mingu Kang and Jaesik Park. Contragan: Contrastive learn- ing for conditional image generation.NeurIPS, 2020. 1

    2020

  16. [24]

    Rebooting acgan: Auxiliary classifier gans with stable training

    Minguk Kang, Woohyeon Shim, Minsu Cho, and Jaesik Park. Rebooting acgan: Auxiliary classifier gans with stable training. InNeural Information Processing Systems, 2021

    2021

  17. [25]

    Studio- gan: a taxonomy and benchmark of gans for image synthesis

    Minguk Kang, Joonghyuk Shin, and Jaesik Park. Studio- gan: a taxonomy and benchmark of gans for image synthesis. IEEE Transactions on Pattern Analysis and Machine Intelli- gence, 2023. 1

    2023

  18. [26]

    Scaling up GANs for Text-to-Image Synthesis.CVPR, 2023

    Minguk Kang, Jun-Yan Zhu, Richard Zhang, Jaesik Park, Eli Shechtman, Sylvain Paris, and Taesung Park. Scaling up GANs for Text-to-Image Synthesis.CVPR, 2023. 5, 6, 1, 2

    2023

  19. [27]

    Progressive growing of gans for improved quality, stability, and variation.ICLR, 2018

    Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gans for improved quality, stability, and variation.ICLR, 2018. 5, 7, 1

    2018

  20. [28]

    Text2video-zero: Text- to-image diffusion models are zero-shot video generators

    Levon Khachatryan, Andranik Movsisyan, Vahram Tade- vosyan, Roberto Henschel, Zhangyang Wang, Shant Navasardyan, and Humphrey Shi. Text2video-zero: Text- to-image diffusion models are zero-shot video generators. ICCV, 2023. 1

    2023

  21. [29]

    Token fusion: Bridging the gap between token pruning and token merging

    Minchul Kim, Shangqian Gao, Yen-Chang Hsu, Yilin Shen, and Hongxia Jin. Token fusion: Bridging the gap between token pruning and token merging. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1383–1392, 2024. 3

    2024

  22. [30]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A Method for Stochastic Optimization.ICLR, 2015. 6, 1

    2015

  23. [31]

    Multi-concept customization of text-to-image diffusion

    Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1931–1941, 2023. 1

    1931

  24. [32]

    Improved precision and recall met- ric for assessing generative models.Advances in neural in- formation processing systems, 32, 2019

    Tuomas Kynk ¨a¨anniemi, Tero Karras, Samuli Laine, Jaakko Lehtinen, and Timo Aila. Improved precision and recall met- ric for assessing generative models.Advances in neural in- formation processing systems, 32, 2019. 5, 1

    2019

  25. [33]

    Distri- fusion: Distributed parallel inference for high-resolution dif- fusion models

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

    2024

  26. [34]

    Faster diffusion: Rethinking the role of unet encoder in diffusion models

    Senmao Li, Taihang Hu, Joost van de Weijer, Fahad Shahbaz Khan, Tao Liu, Linxuan Li, Shiqi Yang, Yaxing Wang, Ming- Ming Cheng, and Jian Yang. Faster diffusion: Rethinking the role of unet encoder in diffusion models. InAdvances in Neural Information Processing Systems, 2024. ...

    2024

  27. [35]

    Styledif- fusion: Prompt-embedding inversion for text-based editing

    Senmao Li, Joost van de Weijer, Taihang Hu, Fahad Shahbaz Khan, Qibin Hou, Yaxing Wang, and Jian Yang. Styledif- fusion: Prompt-embedding inversion for text-based editing. Computational Visual Media Conference, 2024. 1

    2024

  28. [36]

    Q-diffusion: Quantizing diffusion models

    Xiuyu Li, Yijiang Liu, Long Lian, Huanrui Yang, Zhen Dong, Daniel Kang, Shanghang Zhang, and Kurt Keutzer. Q-diffusion: Quantizing diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vi- sion, pages 17535–17545, 2023. 3

    2023

  29. [37]

    Sdxl- lightning: Progressive adversarial diffusion distillation

    Shanchuan Lin, Anran Wang, and Xiao Yang. Sdxl- lightning: Progressive adversarial diffusion distillation. arXiv preprint arXiv:2402.13929, 2024. 2, 3, 4

    2024 arXiv

  30. [38]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceeding...

    2014

  31. [39]

    One-prompt-one-story: Free-lunch consistent text-to-image generation using a single prompt

    Tao Liu, Kai Wang, Senmao Li, Joost van de Weijer, Fa- had Shahbaz Khan, Shiqi Yang, Yaxing Wang, Jian Yang, and Ming-Ming Cheng. One-prompt-one-story: Free-lunch consistent text-to-image generation using a single prompt. CVPR, 2025. 1

    2025

  32. [40]

    Instaflow: One step is enough for high-quality diffusion-based text-to-image generation.ICLR, 2024

    Xingchao Liu, Xiwen Zhang, Jianzhu Ma, Jian Peng, and Qiang Liu. Instaflow: One step is enough for high-quality diffusion-based text-to-image generation.ICLR, 2024. 3, 5, 6, 7, 8, 1, 2

    2024

  33. [41]

    Token caching for diffusion transformer accel- eration.arXiv preprint arXiv:2409.18523, 2024

    Jinming Lou, Wenyang Luo, Yufan Liu, Bing Li, Xinmiao Ding, Weiming Hu, Jiajiong Cao, Yuming Li, and Chen- guang Ma. Token caching for diffusion transformer accel- eration.arXiv preprint arXiv:2409.18523, 2024. 3

    2024

  34. [42]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787,

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongxuan Li, and Jun Zhu. Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps.Advances in Neural Information Processing Systems, 35:5775–5787,

  35. [43]

    Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models.arXiv preprint arXiv:2211.01095, 2022

    Cheng Lu, Yuhao Zhou, Fan Bao, Jianfei Chen, Chongx- uan Li, and Jun Zhu. Dpm-solver++: Fast solver for guided sampling of diffusion probabilistic models.arXiv preprint arXiv:2211.01095, 2022. 1, 3

    2022 arXiv

  36. [44]

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

    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, 2023. 1, 2, 3, 5, 6, 7, 8

    2023 arXiv

  37. [45]

    Lcm-lora: A universal stable-diffusion acceleration module.arXiv preprint arXiv:2311.05556, 2023

    Simian Luo, Yiqin Tan, Suraj Patil, Daniel Gu, Patrick von Platen, Apolin´ario Passos, Longbo Huang, Jian Li, and Hang Zhao. Lcm-lora: A universal stable-diffusion acceleration module.arXiv preprint arXiv:2311.05556, 2023. 1, 2, 3

    2023 arXiv

  38. [46]

    Diff-instruct: A universal approach for transferring knowledge from pre-trained diffu- sion models.NeurIPS, 36, 2023

    Weijian Luo, Tianyang Hu, Shifeng Zhang, Jiacheng Sun, Zhenguo Li, and Zhihua Zhang. Diff-instruct: A universal approach for transferring knowledge from pre-trained diffu- sion models.NeurIPS, 36, 2023. 3

    2023

  39. [47]

    Videofusion: Decomposed diffusion mod- els for high-quality video generation

    Zhengxiong Luo, Dayou Chen, Yingya Zhang, Yan Huang, Liang Wang, Yujun Shen, Deli Zhao, Jingren Zhou, and Tieniu Tan. Videofusion: Decomposed diffusion mod- els for high-quality video generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogniti...

    2023

  40. [48]

    Learning-to-cache: Accelerating diffusion trans- former via layer caching.NeurIPS, 2024

    Xinyin Ma, Gongfan Fang, Michael Bi Mi, and Xinchao Wang. Learning-to-cache: Accelerating diffusion trans- former via layer caching.NeurIPS, 2024. 3

    2024

  41. [49]

    Deepcache: Accelerating diffusion models for free.CVPR, 2024

    Xinyin Ma, Gongfan Fang, and Xinchao Wang. Deepcache: Accelerating diffusion models for free.CVPR, 2024. 1, 3

    2024

  42. [50]

    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. InProceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, pages 14297–14306, 2023. 3

    2023

  43. [51]

    T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models.AAAI, 2023

    Chong Mou, Xintao Wang, Liangbin Xie, Jian Zhang, Zhon- gang Qi, Ying Shan, and Xiaohu Qie. T2i-adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models.AAAI, 2023. 1

    2023

  44. [52]

    Reliable fidelity and diversity metrics for generative models

    Muhammad Ferjad Naeem, Seong Joon Oh, Youngjung Uh, Yunjey Choi, and Jaejun Yoo. Reliable fidelity and diversity metrics for generative models. InInternational Conference on Machine Learning, pages 7176–7185. PMLR, 2020. 5, 7, 1

    2020

  45. [53]

    Swiftbrush: One-step text-to-image diffusion model with variational score distilla- tion.CVPR, 2024

    Thuan Hoang Nguyen and Anh Tran. Swiftbrush: One-step text-to-image diffusion model with variational score distilla- tion.CVPR, 2024. 2, 3, 4, 5, 6, 7, 8, 1

    2024

  46. [54]

    Jour- neyDB: A Benchmark for Generative Image Understanding

    Junting Pan, Keqiang Sun, Yuying Ge, Hao Li, Haodong Duan, Xiaoshi Wu, Renrui Zhang, Aojun Zhou, Zipeng Qin, Yi Wang, Jifeng Dai, Yu Qiao, and Hongsheng Li. Jour- neyDB: A Benchmark for Generative Image Understanding. NeurIPS, 2023. 6, 1

    2023

  47. [55]

    Deep equilibrium approaches to diffusion models

    Ashwini Pokle, Zhengyang Geng, and Zico Kolter. Deep equilibrium approaches to diffusion models. InProceedings of the 36th International Conference on Neural Information Processing Systems, Red Hook, NY , USA, 2024. Curran As- sociates Inc. 3

    2024

  48. [56]

    Barron, and Ben Milden- hall

    Ben Poole, Ajay Jain, Jonathan T. Barron, and Ben Milden- hall. DreamFusion: Text-to-3D using 2D Diffusion.ICLR,

  49. [57]

    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. NeurIPS, 2024. 2, 3

    2024

  50. [58]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022. 1, 2, 4, 5, 6, 8, 3

    2022

  51. [59]

    U- net: Convolutional networks for biomedical image segmen- tation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U- net: Convolutional networks for biomedical image segmen- tation. InMedical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Par...

    2015

  52. [60]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 22500–...

    2023

  53. [61]

    Pho- torealistic text-to-image diffusion models with deep lan- guage understanding.NeurIPS, 2022

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, Burcu Karagol Ayan, S Sara Mahdavi, Rapha Gontijo Lopes, Tim Salimans, Tim Salimans, Jonathan Ho, David J Fleet, and Mohammad Norouzi. Pho- torealistic text-to-imag...

    2022

  54. [62]

    Progressive distillation for fast sampling of diffusion models.ICLR, 2022

    Tim Salimans and Jonathan Ho. Progressive distillation for fast sampling of diffusion models.ICLR, 2022. 1, 3

    2022

  55. [63]

    StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis.International Conference on Machine Learning, 2023

    Axel Sauer, Tero Karras, Samuli Laine, Andreas Geiger, and Timo Aila. StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesis.International Conference on Machine Learning, 2023. 5

    2023

  56. [64]

    Adversarial diffusion distillation.ECCV, 2024

    Axel Sauer, Dominik Lorenz, Andreas Blattmann, and Robin Rombach. Adversarial diffusion distillation.ECCV, 2024. 1, 2, 3, 5, 6, 7, 8, 4

    2024

  57. [65]

    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

  58. [66]

    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 vi- sion and pattern recognition, pages 1972–1981, 2023. 1, 3

    1972

  59. [67]

    Denois- ing diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing diffusion implicit models. InInternational Conference on Learning Representations, 2021. 1, 3, 4, 5, 8

    2021

  60. [68]

    Score-based generative modeling through stochastic differential equa- tions.ICLR, 2021

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Ab- hishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equa- tions.ICLR, 2021. 1

    2021

  61. [69]

    Consistency models

    Yang Song, Prafulla Dhariwal, Mark Chen, and Ilya Sutskever. Consistency models. InInternational Conference on Machine Learning, pages 32211–32252. PMLR, 2023. 1, 3

    2023

  62. [70]

    F3- pruning: A training-free and generalized pruning strategy to- wards faster and finer text-to-video synthesis

    Sitong Su, Jianzhi Liu, Lianli Gao, and Jingkuan Song. F3- pruning: A training-free and generalized pruning strategy to- wards faster and finer text-to-video synthesis. InProceed- ings of the AAAI Conference on Artificial Intelligence, pages 4961–4969, 2024. 1, 3

    2024

  63. [71]

    Emergent correspondence from image diffusion.NeurIPS, 2023

    Luming Tang, Menglin Jia, Qianqian Wang, Cheng Perng Phoo, and Bharath Hariharan. Emergent correspondence from image diffusion.NeurIPS, 2023. 2, 5

    2023

  64. [72]

    Plug-and-play diffusion features for text-driven image-to-image translation

    Narek Tumanyan, Michal Geyer, Shai Bagon, and Tali Dekel. Plug-and-play diffusion features for text-driven image-to-image translation. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 1921–1930, 2023. 2, 5

    1921

  65. [73]

    Phased consistency model.NeurIPS, 2024

    Fu-Yun Wang, Zhaoyang Huang, Alexander William Bergman, Dazhong Shen, Peng Gao, Michael Lingelbach, Keqiang Sun, Weikang Bian, Guanglu Song, Yu Liu, et al. Phased consistency model.NeurIPS, 2024. 3, 2

    2024

  66. [74]

    Yeh, and Greg Shakhnarovich

    Haochen Wang, Xiaodan Du, Jiahao Li, Raymond A. Yeh, and Greg Shakhnarovich. Score Jacobian Chaining: Lifting Pretrained 2D Diffusion Models for 3D Generation.CVPR,

  67. [75]

    Pipefusion: Displaced patch pipeline parallelism for in- ference of diffusion transformer models.arXiv preprint arXiv:2405.14430, 2024

    Jiannan Wang, Jiarui Fang, Aoyu Li, and PengCheng Yang. Pipefusion: Displaced patch pipeline parallelism for in- ference of diffusion transformer models.arXiv preprint arXiv:2405.14430, 2024. 3

    2024 arXiv

  68. [76]

    Dynamic prompt learning: Ad- dressing cross-attention leakage for text-based image edit- ing.NeurIPS, 2023

    Kai Wang, Fei Yang, Shiqi Yang, Muhammad Atif Butt, and Joost van de Weijer. Dynamic prompt learning: Ad- dressing cross-attention leakage for text-based image edit- ing.NeurIPS, 2023. 1

    2023

  69. [77]

    Multi-class textual-inversion secretly yields a semantic-agnostic classifier

    Kai Wang, Fei Yang, Bogdan Raducanu, and Joost van de Weijer. Multi-class textual-inversion secretly yields a semantic-agnostic classifier. InProceedings of the IEEE Workshop on Applications of Computer Vision, 2025. 1

    2025

  70. [78]

    ProlificDreamer: High-Fidelity and Diverse Text-to-3D Generation with Variational Score Distillation.NeurIPS, 2023

    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.NeurIPS, 2023. 4, 5

    2023

  71. [79]

    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

  72. [80]

    One-step diffusion with distribution matching distillation

    Tianwei Yin, Micha ¨el Gharbi, Richard Zhang, Eli Shecht- man, Fredo Durand, William T Freeman, and Taesung Park. One-step diffusion with distribution matching distillation. In CVPR, pages 6613–6623, 2024. 3

    2024

  73. [81]

    Scaling autoregres- sive models for content-rich text-to-image generation.arXiv preprint arXiv:2206.10789, 2022

    Jiahui Yu, Yuanzhong Xu, Jing Yu Koh, Thang Luong, Gun- jan Baid, Zirui Wang, Vijay Vasudevan, Alexander Ku, Yin- fei Yang, Burcu Karagol Ayan, et al. Scaling autoregres- sive models for content-rich text-to-image generation.arXiv preprint arXiv:2206.10789, 2022. 5, 1

    2022 arXiv

  74. [82]

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

    Dingkun Zhang, Sijia Li, Chen Chen, Qingsong Xie, and Haonan Lu. Laptop-diff: Layer pruning and normalized dis- tillation for compressing diffusion models.arXiv preprint arXiv:2404.11098, 2024. 3

    2024 arXiv

  75. [83]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 2023

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Luisa Pola- nia Cabrera, Varun Jampani, Deqing Sun, and Ming-Hsuan Yang. A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence.NeurIPS, 2023. 5

    2023

  76. [84]

    Adding conditional control to text-to-image diffusion models, 2023

    Lvmin Zhang and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models, 2023. 1, 5

    2023

  77. [85]

    Fast sampling of diffusion models via operator learning

    Hongkai Zheng, Weilie Nie, Arash Vahdat, Kamyar Aziz- zadenesheli, and Anima Anandkumar. Fast sampling of diffusion models via operator learning. InProceedings of the 40th International Conference on Machine Learning. JMLR.org, 2023. 3

    2023

  78. [86]

    StudioGAN

    Jianbin Zheng, Minghui Hu, Zhongyi Fan, Chaoyue Wang, Changxing Ding, Dacheng Tao, and Tat-Jen Cham. Trajectory consistency distillation.arXiv preprint arXiv:2402.19159, 2024. 1, 3, 2 One-Way Ticket : Time-Independent Unified Encoder for Distilling Text-to-Image Diffusion Mode...

    2024 arXiv

  79. [2024]

    1, 2, 3, 4, 5, 6, 7, 8

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.