Rethinking Pixel Mean Flows via Interval Denoiser

Aibek Alanov; Alexander Korotin; Alexander Zaytsev; Dmitry Baranchuk

REVIEW 2 major objections 5 minor 15 references

Rethinking Pixel Mean Flows via Interval Denoiser

T0 review · 2 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper derives an exact interval-denoiser identity from the flow-matching ODE and uses it to reach 4.55 FID in one pixel-space step.

desk verdict The Interval Denoiser identity is real and the FID results are strong, but the manifold claim is overreaching; the paper deserves review with revisions. read the letter →

arxiv 2608.04818 v1 pith:6X7CSFI7 submitted 2026-08-05 cs.CV

Alexander Zaytsev , Dmitry Baranchuk , Alexander Korotin , Aibek Alanov This is my paper

classification cs.CV

keywords intervaldenoiserpixel-spacegenerationfew-stepsamplingflowmatchinglatent-freegenerativemodelsdenoised-imagemanifoldresidualclippingtime-samplingcurriculum

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 proposes the Interval Denoiser, a prediction target for few-step, latent-free image generation that is derived directly from the flow-matching ODE rather than introduced by hand. The central claim is that intermediate trajectory states can be mapped exactly to a weighted average of denoiser outputs, and that this average stays on the low-dimensional manifold of clean images, which makes direct pixel-space regression tractable. The paper further claims that this exact formulation avoids the biased gradients that arise when prior methods substitute an image prediction into a velocity objective. Trained from scratch on ImageNet 256x256 without perceptual losses, the resulting model achieves an FID of 4.55 in one step and 3.98 in two steps.

What carries the argument

The central object is the Interval Denoiser $X(z_t,r,t)=\frac{t r}{t-r}\int_r^t \frac{x(z_\tau,\tau)}{\tau^2}\,d\tau$, a normalized weighted aggregation of instantaneous denoising predictions along a flow trajectory. The Interval Denoiser Identity $X+\frac{t(t-r)}{r}\frac{dX}{dt}=x$ turns this aggregate into a regression target: a network predicts $X$ directly, while the time derivative is computed from the network itself via a stop-grad Jacobian-vector product, so no spatial prediction is hidden inside the stop-gradient. This identity carries the argument because it connects the tractable on-manifold target $X$ to the exact ODE state update and exposes the extra terms that bias prior pixel-space objectives.

What would settle it

On a synthetic dataset whose clean images form a known non-convex set, train the Interval Denoiser objective and check whether $X(z_t,r,t)$ lies within the set for all intervals; any interval where it falls outside the set refutes the manifold claim.

Watch

Extended reading notes

Core claim

The paper's central discovery is the Interval Denoiser identity $X(z_t,r,t)+\frac{t(t-r)}{r}\frac{d}{dt}X(z_t,r,t)=x(z_t,t)$, an exact analytical relation between the interval-aggregate prediction $X$ and the instantaneous denoiser $x$, obtained by integrating the flow-matching ODE. It shows that $X$ is a normalized weighted average of denoiser estimates that all target the same clean image, so it belongs to the same low-dimensional denoised-image manifold at every interval; this is what makes pixel-space few-step learning tractable. The derivation also proves that the algebraic substitution used by prior pixel-space mean-flow methods is a formal consequence of image-space prediction, and shows that those methods' losses trap spatial prediction terms inside stop-gradients, producing biased updates. The proposed objective isolates the pure time derivative, and with residual clipping and a two-phase time-sampling curriculum, a pixel-space model trained from scratch achieves 4.55 FID at one function evaluation and 3.98 at two.

Load-bearing premise

The load-bearing premise is that a weighted average of denoised estimates of the same image is itself a clean image; this is not guaranteed, because averaging points on a non-convex image manifold can leave the manifold.

Editorial extensions

If this is right

The exact identity gives pixel-space few-step models a principled training signal, so one-step and two-step sampling no longer depend on an ad hoc substitution into a velocity objective.
Because the formulation isolates the pure time derivative, parameter updates follow the analytical gradient rather than a biased approximation, which should improve training stability and sample quality at long intervals.
The framework formally recovers the structural form of consistency-trajectory models, implying that the preconditioned parameterization used there is not a separate design choice but a consequence of exact interval integration.
The ablation results indicate that residual clipping and the time-sampling curriculum are needed for wide-interval training; clipping improves one-step FID from 9.78 to 9.25 and the curriculum further reduces it to 7.55 on the base architecture.
Scaling the model and training budget brings pixel-space few-step generation to 4.55 FID (1-NFE) and 3.98 FID (2-NFE) on ImageNet 256x256, closing much of the gap to latent-space models without a tokenizer.

Reading between the lines

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

A direct test of the manifold claim would be to train on synthetic data with a known non-convex clean-image set, such as two separated clusters, and measure whether the predicted $X$ stays on that set; if it does not, the tractability argument would need revision.
The same identity should transfer to other flow-matching modalities, such as audio or video, wherever the denoiser is an expectation and the generation trajectory is linear, because the derivation uses only the linear flow ODE.
The paper's diagnosis of biased stop-gradient losses suggests a broader principle: any few-step distillation objective that hides spatial prediction terms inside a stop-gradient should be re-derived from the ODE before being used as a training loss.
If the manifold property holds, the framework could also serve as a diagnostic tool for measuring how far a trained few-step sampler drifts off the clean-image manifold as intervals widen.

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.

Desk Editor's Note

read the letter

Dear colleague,

Quick take: this paper deserves a serious referee, though not necessarily acceptance as-is. The central identity is real, the FID numbers are good, and the bias analysis is a genuine contribution. The weak spots are an unsupported manifold claim and some overstated optimization language.

What's new: they derive from the flow-matching ODE a closed-form expression for an interval denoiser X(z_t,r,t) that averages denoiser outputs along the trajectory. This identity formally reproduces the empirical substitution used in Pixel MeanFlow and the CTM parameterization. They also demonstrate that pMF's loss, when written in terms of an image-space network, has extra spatial prediction terms trapped inside the stop-gradient operator, and their own loss avoids that. That is a clean and useful observation. Prop. 2 confirms the equivalence to the MeanFlow objective with a scaling factor.

The empirical results are credible. pID-L/16 hits 4.55 FID at one NFE without perceptual losses, versus 8.71 for the pMF-B/16 baseline, and the ablations show residual clipping and the time-sampling curriculum each contribute substantially. They document failed experiments, which adds trust. The comparison to latent-space methods is fair in the sense that they clearly separate the settings.

Now the soft spots. The manifold claim: Prop. 1 shows X is a weighted average of conditional expectations E[x0|z_tau]. The paper then asserts that because each term estimates the same clean image, the average also lies in the low-dimensional manifold of denoised images. That does not follow. A convex combination of points in a non-convex set can leave the set. This is not a minor wording issue: the abstract says the prediction 'is shown to reside' on the manifold, and this is used to justify pixel-space tractability. The practical method may still work without this theorem, but the claim as stated is unproven. The optimization claim is softer: the loss replaces the true denoiser with x0 and stops the JVP, so the gradient is exact for the surrogate loss, not for the original ODE identity. That's a wording overreach rather than a technical flaw.

Also note: no code is released, and the headline FID depends on several tuned hyperparameters (logit-normal parameters, transition epoch, residual clipping bound, CFG intervals). That's normal for this area, but it tempers how much we can judge from the table alone.

Bottom line: the core algebra is correct, the empirical recipe is effective, and the bias analysis is a step forward. I'd send this to peer review and ask the authors to either prove or soften the manifold claim, and to phrase the optimization guarantees more carefully. It would be a useful citation for anyone working on few-step pixel-space generation.

Regards, [no name]

Referee Report

2 major / 5 minor

Summary. The paper proposes the Interval Denoiser (pID), a latent-free pixel-space few-step generative model. It derives an identity from the flow matching ODE (Eq. 14) that relates the integral of instantaneous denoisers over an interval, X(z_t,r,t), to the instantaneous denoiser x(z_t,t). This identity is used to define a training objective with a stop-gradient Jacobian-vector product and a sampling update (Eq. 18). The authors claim that X resides on a low-dimensional denoised-image manifold, making pixel-space regression tractable, and that their loss ensures exact first-order optimization. They introduce residual clipping and a two-phase time-sampling curriculum. On ImageNet 256x256, pID-L/16 achieves FID 4.55 at 1-NFE and 3.98 at 2-NFE without perceptual losses, outperforming the pixel MeanFlow (pMF) baseline without perceptual losses.

Significance. The Interval Denoiser identity (Eq. 14) is a clean and exact analytical result that connects several existing formulations: it formally derives the pMF substitution and recovers the CTM decoder parameterization. The experimental results are strong, with careful ablations of residual clipping and time-sampling curriculum, and the paper honestly documents failed experiments in Appendix D. However, the paper's central theoretical justifications—the manifold claim and the claim of exact first-order optimization—are not convincingly established. The empirical gains over the no-perceptual-loss pMF baseline are meaningful, but the theoretical narrative overstates what is proven. The paper is otherwise reproducible, with detailed hyperparameters and computational budget.

major comments (2)

[Fundamentals of Interval Denoising (Prop. 1 and the 'Generalized Manifold Hypothesis' paragraph)] The claim that the Interval Denoiser X(z_t,r,t) lies on the low-dimensional manifold of denoised images does not follow from Prop. 1. Prop. 1 only proves that X is a weighted average of conditional expectations E[x0|z_tau] along the trajectory. The set of denoised images (or the image manifold) is not convex, so a convex combination of points in that set need not lie in the set. The paper states: 'By Prop. 1, X averages denoiser outputs along a single trajectory, all estimating the same clean image x0. For any interval, the average is again an estimate of x0, so it lies in the same low-dimensional set of denoised images.' This is a non sequitur. The abstract and contributions claim that the prediction 'is shown to reside on a low-dimensional manifold across any time interval,' and this property is used to justify pixel-space tractability. Since the proof is only for the weighted average property, the manifold claim is an unproven assumption, not a theorem. The authors should either provide a proof under additional assumptions (e.g., convexity of the denoised-image set, which is unlikely) or revise the claim to an empirical observation, clearly labeling it as a hypothesis.
[Biased Optimization via Algebraic Substitutions and Eq. 17] The claim of 'exact first-order optimization' is overstated. In Eq. 17, the JVP term is computed with a stop-gradient network copy θ−, so the gradient of the loss with respect to θ is not the gradient of the full objective where the JVP depends on θ. The paper argues that pMF 'traps spatial prediction terms inside the stop-gradient,' but in its own objective, the entire JVP term is frozen. Although the JVP is not a spatial prediction, it still depends on the network through the auxiliary head xθ used in the tangent vector (z−x,0,t). Thus the gradient used in training is the gradient of a surrogate objective, not the exact gradient of the original functional. The statement in Section 'Biased Optimization via Algebraic Substitutions' that 'Applying the stop-gradient to the JVP hides no spatial parameters, ensuring exact first-order optimization' is misleading. The authors should clarify what 'exact' means here (e.g., exact gradient of the surrogate loss) and discuss the relationship between their stop-gradient choice and the pMF bias they criticize.

minor comments (5)

[Table 2] The baseline 'JiT-L/16 (2025)' is listed in Table 2 but does not appear in the references; the authors should provide a citation or remove it.
[Eq. 21 and surrounding text] The notation sg(·) for stop-gradient is used in Eq. 21 but is not formally defined before its first use; please define it in the background or in the notation section.
[Algorithm 1] In Algorithm 1, the line 'x = net(z, t, t)' is ambiguous because the network is otherwise called with (z, r, t); it should be clarified that this is the auxiliary instantaneous-denoiser head evaluated at the boundary r=t.
[Figure 1] The geometry diagram in Figure 1 is hard to parse; the axes and the meaning of the curves (especially the dotted segment and the labels M and 'low dimensional manifold') should be defined explicitly in the caption.
[Experiments / Main Results] The claim of 'state-of-the-art' in the abstract is qualified to models without perceptual losses; this scope should be stated clearly in the main text as well, to avoid the impression of a broader comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Interval Denoiser identity is an exact ODE rearrangement with no fitted parameters, and the reported gains are measured against external FID benchmarks.

full rationale

The central identity (Eq. 14) follows by differentiating the definition Eq. 12, which is itself just the normalized integral form of the exact flow-matching state update Eq. 11; it introduces no fitted constants and does not presuppose the network's output. The training objective Eq. 17 regresses Xθ(zt,r,t)+(t-r)/r·JVP against the ground-truth image x0, the standard unbiased surrogate for the conditional expectation E[x0|zt], so the regression target is external to the model rather than a rename of a fitted parameter. Residual clipping (Eqs. 30-31) does use a stop-gradient self-distillation target, but the paper explicitly presents it as a stabilization device and validates it with ablations against FID, so the empirical claims are not forced by construction. The main theoretical weakness is Proposition 1's jump from 'X is a convex combination of conditional expectations' to 'X lies on the low-dimensional denoised-image manifold': convex combinations of points on a non-convex manifold need not remain on it. That is an unsupported premise (a rigor gap), not a circular reduction, and it does not undermine the algebraic derivation or the external benchmark results. No load-bearing self-citation or imported uniqueness theorem appears; connections to pMF and CTM are explicitly derived equivalences rather than renamed inputs.

Assumptions & free parameters 6 free parameters · 5 assumptions · 1 invented entities

The central claim rests on a few free hyperparameters (curriculum, clipping, inference CFG) and on the unproven manifold hypothesis. The Interval Denoiser itself is a reparameterization, not a new physical entity.

free parameters (6)

Phase I logit-normal sampler (mu=0.8, sigma=0.8) = mu=0.8, sigma=0.8
Selected by FID ablation in Table 1 over mu=0.0.
Phase II uniform sampler with 100% mixing, transition epoch Ts=140 = 100% mixing, Ts=140
Tuned in Table 1; Ts=140 gives best 7.55 FID.
Residual clipping bound = [-1, 1]
Chosen by hand to bound residual variance.
Denominator clipping r_min = 0.05
Prevents divergence as r approaches 0; ablation shows sensitivity, removed by curriculum.
Log objective delta = 0.01
Small constant in log loss, chosen for stability.
Inference CFG scale and intervals = omega=7.0, interval [0.1,0.82] (1-NFE), [0.1,0.74], k=0.8 (2-NFE)
Selected to minimize FID in Appendix E.

assumptions (5)

standard math Flow matching ODE with linear interpolation z_t=(1-t)x0+t*eps defines a valid probability path.
Used in background Eq. 1 and throughout the derivation.
domain assumption A neural network can represent the Interval Denoiser and instantaneous denoiser with sufficient accuracy.
Assumed implicitly by training a finite-capacity ViT; no capacity bound is proven.
domain assumption Stop-gradient JVP with regression to x0 yields a valid and effectively unbiased training objective.
Shared by Improved MeanFlow and CTM; not rigorously justified here, only motivated.
ad hoc to paper Generalized manifold hypothesis: the weighted average X stays on the low-dimensional denoised-image manifold.
Prop. 1 proves only that X is an expectation; the manifold property is asserted, not proven.
domain assumption Residual clipping and the two-phase curriculum do not introduce distributional bias that degrades final sample quality.
Validated empirically only in ablations; no theoretical guarantee.

invented entities (1)

Interval Denoiser X(z_t, r, t) independent evidence
purpose: Prediction target that maps a noisy state at time t to a denoised-image estimate over the interval [r,t], enabling one-step pixel-space sampling.
It is a new named object derived from the ODE, but algebraically equivalent to CTM's data predictor g_theta; its usefulness is evidenced by the FID results and the mathematical identity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Rethinking Pixel Mean Flows via Interval Denoiser." pith.science (2026). https://pith.science/paper/6X7CSFI7

@misc{pith2026260804818,
  author       = {Pith},
  title        = {Pith review of: Rethinking Pixel Mean Flows via Interval Denoiser},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6X7CSFI7}},
  note         = {Machine review of arXiv:2608.04818}
}

read the original abstract

Modern diffusion and flow-based models are increasingly moving toward few-step, latent-free generation to bypass the computational overhead of multi-step sampling and the reconstruction bottlenecks of external autoencoders. We propose the Interval Denoiser, a theoretically rigorous framework for latent-free generation. Derived directly from the flow matching ODE, it establishes an exact analytical mapping for intermediate trajectory states. Unlike prior formulations, our prediction is shown to reside on a low-dimensional manifold across any time interval, making the regression tractable for a network operating directly on pixels. Furthermore, by avoiding empirical algebraic substitutions, our formulation correctly isolates the pure time derivative to prevent biased gradient evaluations and ensure exact first-order optimization. By analyzing this objective, we equip our framework with residual clipping and a time-sampling curriculum, enabling effective long-interval training and improving few-step performance. Trained from scratch on ImageNet 256x256, our model achieves an FID of 4.55 in one step (1-NFE) and 3.98 in two steps (2-NFE) without perceptual losses.

Figures

Figures reproduced from arXiv: 2608.04818 by the authors.

**Figure 2.** Training dynamics and interval sampling analysis. We compare our pID with pMF. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Effect of residual clipping on 1-NFE training. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Analysis of 2-NFE intermediate sampling trajectory states. (a) Intermediate state visualizations after Step 1 ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references · 9 canonical work pages

[5]

Understanding, Accelerating, and Improving MeanFlow Training

Un- derstanding, Accelerating, and Improving MeanFlow Train- ing.arXiv preprint arXiv:2511.19065. Kynkäänniemi,T.;Karras,T.;Aittala,M.;Aila,T.;andLehti- nen, J

work page Pith review arXiv
[6]

arXiv:2511.13720

Back to Basics: Let Denoising Generative Models Denoise. arXiv:2511.13720. Lipman, Y.; Chen, R. T. Q.; Ben-Hamu, H.; Nickel, M.; and Le, M

arXiv
[7]

In International Conference on Learning Representations

Flow Matching for Generative Modeling. In International Conference on Learning Representations. Lu,C.;andSong,Y.2025. Simplifying,StabilizingandScal- ing Continuous-Time Consistency Models. InInternational Conference on Learning Representations. Lu, Y.; Lu, S.; Sun, Q.; Zhao, H.; Jiang, Z.; Wang, X.; Li, T.; Geng, Z.; and He, K

work page 2025
[9]

InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10684–10695

High-Resolution Image Synthesis with La- tent Diffusion Models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10684–10695. Sabour,A.;Fidler,S.;andKreis,K.2025. AlignYourFlow: Scaling Continuous-Time Flow Map Distillation. InAd- vances in Neural Information Processing Systems (NeurIPS). Salimans,T.;Goodfellow,I.;Zarem...

work page 2025
[10]

InACM SIGGRAPH 2022 Conference Proceedings

StyleGAN- XL: Scaling StyleGAN to Large Diverse Datasets. InACM SIGGRAPH 2022 Conference Proceedings. Shi, M.; Wang, H.; Zheng, W.; Yuan, Z.; Wu, X.; Wang, X.; Wan, P.; Zhou, J.; and Lu, J

work page 2022
[11]

Song, Y.; and Dhariwal, P

Latent Diffusion Model Without Variational Autoencoder.arXiv preprint arXiv:2510.15301. Song, Y.; and Dhariwal, P

arXiv
[12]

arXiv:2507.23268

PixNerd: Pixel Neural Field Diffusion. arXiv:2507.23268. Yao, J.; Yang, B.; and Wang, X

arXiv
[13]

Generation:TamingOptimizationDilemmainLatentDiffu- sion Models

Reconstruction vs. Generation:TamingOptimizationDilemmainLatentDiffu- sion Models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Zhang, H.; Siarohin, A.; Menapace, W.; Vasilkovsky, M.; Tulyakov,S.;Qu,Q.;andSkorokhodov,I.2026. AlphaFlow: Understanding and Improving MeanFlow Models. InInter- national Conference on Lear...

work page 2017

Show all 15 references

[14]

directly into training rather than at inference. By sub- stituting our Interval Denoiser into the velocity guidance formula via the identityu= (z t −X)/t, we construct the guided target: xcfg =x 0+ 1− 1 ω Xθ− (zt, t, t|c)−Xθ− (zt, t, t| ∅) . (32) Here,cand∅denote the condition...

2026
[140]

Baselines.Our pMF-B/16 reproduction at 160 epochs (without residual clipping) yields 9.56 FID, matching the value reported by Lu et al. (2026). C Computational Budget Models are trained on a single node with 8 NVIDIA H100 GPUs.pID-B/16requiresaround3daysfor160epochs(576 H100-h...

2026
[2017]

arXiv:1706.02677

Accu- rate, Large Minibatch SGD: Training ImageNet in 1 Hour. arXiv:1706.02677. Heusel, M.; Ramsauer, H.; Unterthiner, T.; Nessler, B.; and Hochreiter, S

arXiv
[2022]

arXiv:2207.12598

Classifier-Free Diffusion Guidance. arXiv:2207.12598. Jabri, A.; Fleet, D.; and Chen, T

arXiv
[2023]

InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition

Scaling Up GANs for Text-to- Image Synthesis. InProceedings of the IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition. Karras,T.;Aittala,M.;Aila,T.;andLaine,S.2022.Elucidat- ingtheDesignSpaceofDiffusion-BasedGenerativeModels. InAdvances in Neural Information Proc...

2022
[2025]

Brock,A.;Donahue,J.;andSimonyan,K.2019

Stochastic Interpolants: A Unifying Framework for Flows and Diffusions.Journal of Machine Learning Research, 26: 1–80. Brock,A.;Donahue,J.;andSimonyan,K.2019. LargeScale GANTrainingforHighFidelityNaturalImageSynthesis. In International Conference on Learning Representations. C...

2019
[2026]

arXiv:2601.22158

One-Step Latent-Free Image Generation with Pixel Mean Flows. arXiv:2601.22158. Peng, Y.; Zhu, K.; Liu, Y.; Wu, P.; Li, H.; Sun, X.; and Wu, F

arXiv

Pith tools

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

[1] [5]

Understanding, Accelerating, and Improving MeanFlow Training

Un- derstanding, Accelerating, and Improving MeanFlow Train- ing.arXiv preprint arXiv:2511.19065. Kynkäänniemi,T.;Karras,T.;Aittala,M.;Aila,T.;andLehti- nen, J

work page Pith review arXiv

[2] [6]

arXiv:2511.13720

Back to Basics: Let Denoising Generative Models Denoise. arXiv:2511.13720. Lipman, Y.; Chen, R. T. Q.; Ben-Hamu, H.; Nickel, M.; and Le, M

arXiv

[3] [7]

In International Conference on Learning Representations

Flow Matching for Generative Modeling. In International Conference on Learning Representations. Lu,C.;andSong,Y.2025. Simplifying,StabilizingandScal- ing Continuous-Time Consistency Models. InInternational Conference on Learning Representations. Lu, Y.; Lu, S.; Sun, Q.; Zhao, H.; Jiang, Z.; Wang, X.; Li, T.; Geng, Z.; and He, K

work page 2025

[4] [9]

InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10684–10695

High-Resolution Image Synthesis with La- tent Diffusion Models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 10684–10695. Sabour,A.;Fidler,S.;andKreis,K.2025. AlignYourFlow: Scaling Continuous-Time Flow Map Distillation. InAd- vances in Neural Information Processing Systems (NeurIPS). Salimans,T.;Goodfellow,I.;Zarem...

work page 2025

[5] [10]

InACM SIGGRAPH 2022 Conference Proceedings

StyleGAN- XL: Scaling StyleGAN to Large Diverse Datasets. InACM SIGGRAPH 2022 Conference Proceedings. Shi, M.; Wang, H.; Zheng, W.; Yuan, Z.; Wu, X.; Wang, X.; Wan, P.; Zhou, J.; and Lu, J

work page 2022

[6] [11]

Song, Y.; and Dhariwal, P

Latent Diffusion Model Without Variational Autoencoder.arXiv preprint arXiv:2510.15301. Song, Y.; and Dhariwal, P

arXiv

[7] [12]

arXiv:2507.23268

PixNerd: Pixel Neural Field Diffusion. arXiv:2507.23268. Yao, J.; Yang, B.; and Wang, X

arXiv

[8] [13]

Generation:TamingOptimizationDilemmainLatentDiffu- sion Models

Reconstruction vs. Generation:TamingOptimizationDilemmainLatentDiffu- sion Models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Zhang, H.; Siarohin, A.; Menapace, W.; Vasilkovsky, M.; Tulyakov,S.;Qu,Q.;andSkorokhodov,I.2026. AlphaFlow: Understanding and Improving MeanFlow Models. InInter- national Conference on Lear...

work page 2017