pith. sign in

arxiv: 2606.00094 · v2 · pith:3RZGA262new · submitted 2026-05-25 · 💻 cs.CV · cs.AI

Diffusion Image Generation with Explicit Modeling of Data Manifold Geometry

Duoduo Xue , Zhiyu Zhu , Junhui Hou This is my paper

Pith reviewed 2026-06-29 22:42 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords diffusion modelsimage generationdata manifoldpatch tokenizationsoft top-k aggregationImageNetFID scorehigh-frequency embeddings
0
0 comments X

The pith

Integrating discrete patch tokenization into continuous diffusion score functions explicitly models data manifold geometry for better image generation.

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

The paper establishes that discrete patch tokenization can be integrated into the score function of a continuous diffusion model to explicitly capture the geometry of the underlying data manifold. This hybrid method uses soft top-k aggregation to enable end-to-end differentiable training while retaining the structural quantification of discrete tokens and the parallel flexibility of continuous diffusion. Dual-branch high-frequency embeddings and multi-stage transition sampling further support the approach by addressing transformer spectral bias and adjusting inference dynamically. Experiments show this yields lower FID scores on ImageNet-256x256 than baselines, including a 1.95 FID with 715M parameters that surpasses a 3.1B-parameter model. A sympathetic reader would care because it points to a more compact parameterization of the dense low-dimensional space from which data points are sampled.

Core claim

MIND explicitly models manifold geometry by integrating discrete patch tokenization into the score function of a continuous diffusion model, enabled by soft top-k aggregation for end-to-end training, dual-branch high-frequency feature embeddings, and multi-stage transition sampling, resulting in an FID of 1.95 on ImageNet-256×256 with 715M parameters while surpassing larger models such as LlamaGen-3B.

What carries the argument

Soft top-k aggregation mechanism that integrates discrete patch tokenization into the continuous diffusion score function to enable explicit manifold geometry modeling.

If this is right

  • The base model achieves an FID of 22.73 without guidance after 80 epochs, nearly halving the 43.47 FID of the vanilla DiT-B/2 baseline.
  • The method reduces FID by 15.95 on average compared with DiT baselines and by 9.06 compared with SiT baselines.
  • MIND-B with 130M parameters reaches an FID of 2.06 with guidance, surpassing the 3.1B-parameter LlamaGen-3B.
  • MIND-XL with 715M parameters further reduces the FID to 1.95.

Where Pith is reading between the lines

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

  • The hybrid discrete-continuous structure may extend to other high-dimensional generative tasks where both structural tokens and flexible sampling are useful.
  • Multi-stage transition sampling could be tested in non-image domains such as audio waveform generation to handle timestep-dependent transitions.
  • Controlled ablations isolating the manifold-modeling component would clarify whether geometry awareness improves sample diversity in addition to fidelity.
  • Applying the same tokenization integration at higher resolutions could test whether the parameter efficiency scales beyond 256x256 images.

Load-bearing premise

The soft top-k aggregation preserves sufficient information from the discrete tokens so that the manifold geometry modeling, rather than other implementation details, drives the reported FID improvements.

What would settle it

Train an otherwise identical model without the discrete patch tokenization and soft top-k components on the same dataset and epochs, then measure whether the FID rises substantially above 1.95.

Figures

Figures reproduced from arXiv: 2606.00094 by Duoduo Xue, Junhui Hou, Zhiyu Zhu.

Figure 1
Figure 1. Figure 1: Illustration of the proposed MIND, an image generation framework with explicit modeling of data manifold geometry. 3 Proposed Method The low-dimensional manifold hypothesis is widely adopted in the theoretical analysis of image generation with diffusion models [2,10,42,51,63,65], but has not been explicitly utilized in modern generative models. In this paper, we leverage the image tokenization method as th… view at source ↗
Figure 2
Figure 2. Figure 2: Continuous (top) and discrete projection (bot￾tom). The plot shows the LPIPS distance between the reconstructed and original images during optimization. Solid lines represent the models with manifold con￾straints, whereas dashed lines indicate the unconstrained reconstruction. To mitigate these challenges, we employ dis￾crete tokenization to quantize the manifold struc￾ture, an approach that significantly … view at source ↗
Figure 3
Figure 3. Figure 3: Visual comparison between our MIND-B and SiT-B/2 with cfg=2.0. classifier-free guidance scales. Under the small-scale setting (∼35M parameters), the proposed MIND-S significantly outperforms the baseline models. Specifically, at cfg = 1.5, MIND-S improves the FID by 22.09 and 13.29 compared with DiT-S/2 and SiT-S/2, respectively. MIND-S surpasses eMIGM-S with only one-third of the parameters. Scaling up to… view at source ↗
Figure 4
Figure 4. Figure 4: Examples of visual results from our MIND-B with cfg=4.0. Furthermore, our method maintains a distinct advantage over continuous architectures. Specifically, compared with its baseline DiT-XL/2 [49], MIND improves the generation ability from FID=2.27 to 1.95. The proposed MIND using a stronger tokenizer named GigaTok [30], outperforms the 2B-parameter SimDiff [27] and SiT-XL/2 [44]. Compared with recent mas… view at source ↗
Figure 5
Figure 5. Figure 5: Examples of visual results from our MIND-B with FID=2.21 [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Examples of visual results from our one-step MIND-B-G [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Convergence speed [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Examples of visual results from our MIND-B-G for text-to-image generation [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
read the original abstract

Image generative models aim to sample data points from the underlying data manifold, a task that requires learning and decoding a dense, low-dimensional, and compact parameterization space. To achieve this, we propose the Data Manifold-aware Image diffusioN moDel (MIND), a novel framework that explicitly models manifold geometry by integrating discrete patch tokenization into the score function of a continuous diffusion model. This approach successfully leverages both the structural quantification capabilities of discrete tokens and the parallel generation flexibility of continuous diffusion. Moreover, we enable end-to-end differentiable training via a novel soft top-$k$ aggregation mechanism and introduce dual-branch high-frequency feature embedding layers to alleviate the spectral bias of transformer backbones on low-dimensional inputs. Furthermore, for inference, we design a multi-stage transition sampling scheme that dynamically adjusts the sampling scheme based on timestep. Extensive experiments on ImageNet 256$\times$256 demonstrate the effectiveness of MIND. After 80-epoch training, our base model achieves an FID of 22.73 without guidance, nearly halving the 43.47 FID of the vanilla DiT-B/2 baseline. The proposed method reduces FID by 15.95 and 9.06 on average compared with the baselines DiT and SiT, respectively. For image generation on ImageNet-256$\times$256 with guidance, the proposed MIND-B with only 130M parameters achieves an FID of 2.06, superpassing the LlamaGen-3B with 3.1B parameters. The proposed MIND-XL with 715M parameters further reduces the FID to 1.95. Our MIND introduces a fresh perspective on diffusion-based image generation, paving the way for future research and innovation in this community. The code will be publicly available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper proposes the Data Manifold-aware Image diffusioN moDel (MIND), a hybrid framework that integrates discrete patch tokenization into the score function of a continuous diffusion model via a soft top-k aggregation mechanism, combined with dual-branch high-frequency feature embeddings and multi-stage transition sampling. It claims this explicitly models data manifold geometry and reports strong empirical results on ImageNet 256×256, including an FID of 22.73 for the base model after 80 epochs (vs. 43.47 for DiT-B/2), average FID reductions of 15.95 vs. DiT and 9.06 vs. SiT, and guided FID scores of 2.06 (MIND-B, 130M params) and 1.95 (MIND-XL, 715M params) that surpass much larger baselines like LlamaGen-3B.

Significance. If the performance gains are reproducible and attributable to the manifold-geometry modeling rather than unablated implementation details, the hybrid discrete-continuous approach could provide a more parameter-efficient path for diffusion-based generation. The empirical comparisons to DiT/SiT and larger autoregressive models are potentially impactful for the field, but the lack of mechanistic evidence or ablations limits the theoretical advance.

major comments (3)
  1. [Abstract] Abstract: The central claim that the method 'explicitly models manifold geometry' by integrating discrete patch tokenization into the continuous score function lacks any equation, derivation, or section defining how the soft top-k aggregation or token integration enforces geometric properties (e.g., curvature, geodesics, or structural constraints); without this, it is unclear whether the discrete component contributes beyond the dual-branch embeddings and sampling scheme.
  2. [Abstract] Abstract: The reported FID values (22.73 after 80 epochs for base model; 1.95 for MIND-XL) and comparisons (e.g., vs. DiT-B/2 at 43.47, LlamaGen-3B) provide no experimental details, error bars, baseline implementation specifications, training hyperparameters, or ablation studies, rendering the performance claims unverifiable and the attribution to manifold modeling unsupported.
  3. [Abstract] Abstract: The soft top-k aggregation is presented as enabling end-to-end differentiability while preserving 'structural quantification capabilities of discrete tokens,' but no analysis or equation demonstrates that it avoids collapse to a smoothed continuous signal (high-entropy selection) that would undermine the discrete manifold modeling.
minor comments (1)
  1. [Abstract] Abstract contains minor language issues including 'superpassing' (should be 'surpassing') and inconsistent capitalization in the model name 'diffusioN moDel'.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each major comment point by point below, clarifying the manuscript's content and indicating where revisions will strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that the method 'explicitly models manifold geometry' by integrating discrete patch tokenization into the continuous score function lacks any equation, derivation, or section defining how the soft top-k aggregation or token integration enforces geometric properties (e.g., curvature, geodesics, or structural constraints); without this, it is unclear whether the discrete component contributes beyond the dual-branch embeddings and sampling scheme.

    Authors: The abstract summarizes the high-level contribution. The method section of the manuscript defines the integration of discrete patch tokenization into the score function via soft top-k aggregation and explains its role in capturing manifold structure through structural quantification. To directly address the request for explicit definitions, we will add a dedicated subsection with equations and a short derivation showing how the token integration imposes local geometric constraints on the diffusion process. revision: yes

  2. Referee: [Abstract] Abstract: The reported FID values (22.73 after 80 epochs for base model; 1.95 for MIND-XL) and comparisons (e.g., vs. DiT-B/2 at 43.47, LlamaGen-3B) provide no experimental details, error bars, baseline implementation specifications, training hyperparameters, or ablation studies, rendering the performance claims unverifiable and the attribution to manifold modeling unsupported.

    Authors: The abstract condenses the key results. Full experimental details, including training hyperparameters, baseline re-implementation specifications, ablation studies, and dataset protocols, appear in Sections 4 and 5. We will expand the abstract's result paragraph with references to these sections and add error bars computed over multiple random seeds plus explicit baseline configuration tables in the revision. revision: yes

  3. Referee: [Abstract] Abstract: The soft top-k aggregation is presented as enabling end-to-end differentiability while preserving 'structural quantification capabilities of discrete tokens,' but no analysis or equation demonstrates that it avoids collapse to a smoothed continuous signal (high-entropy selection) that would undermine the discrete manifold modeling.

    Authors: The manuscript introduces the soft top-k mechanism for differentiability. We will include a new paragraph with the explicit formulation of the aggregation operator and an entropy analysis demonstrating that the temperature schedule keeps selection entropy sufficiently low to retain discrete-like behavior, thereby supporting the manifold-modeling claim. This analysis will be added to the method section. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper introduces an architectural framework (MIND) combining discrete patch tokenization, soft top-k aggregation, dual-branch embeddings, and multi-stage sampling within a diffusion score function. No equations, uniqueness theorems, or first-principles derivations are presented that reduce by construction to fitted parameters, self-citations, or renamed inputs. All reported outcomes consist of empirical FID comparisons against external baselines (DiT, SiT, LlamaGen) after fixed training epochs, with no load-bearing self-referential steps or ansatz smuggling. The work is self-contained as an empirical model proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5858 in / 1109 out tokens · 39036 ms · 2026-06-29T22:42:18.981051+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

82 extracted references · 12 canonical work pages · 6 internal anchors

  1. [1]

    In: NeurIPS

    Austin, J., Johnson, D.D., Ho, J., Tarlow, D., van den Berg, R.: Structured denoising diffusion models in discrete state-spaces. In: NeurIPS. pp. 17981–17993 (2021)

    2021

  2. [2]

    arXiv preprint arXiv:2409.18804 (2024)

    Azangulov, I., Deligiannidis, G., Rousseau, J.: Convergence of diffusion models under the manifold hypothesis in high-dimensions. arXiv preprint arXiv:2409.18804 (2024)

    work page arXiv 2024

  3. [3]

    In: ICLR (2022)

    Bao, F., Li, C., Zhu, J., Zhang, B.: Analytic-dpm: an analytic estimate of the optimal reverse variance in diffusion probabilistic models. In: ICLR (2022)

    2022

  4. [4]

    In: CVPR

    Blattmann, A., Rombach, R., Ling, H., Dockhorn, T., Kim, S.W., Fidler, S., Kreis, K.: Align your latents: High-resolution video synthesis with latent diffusion models. In: CVPR. pp. 22563–22575 (2023)

    2023

  5. [5]

    In: ICLR (2019)

    Brock, A., Donahue, J., Simonyan, K.: Large scale GAN training for high fidelity natural image synthesis. In: ICLR (2019)

    2019

  6. [6]

    In: CVPR (2022)

    Chang, H., Zhang, H., Jiang, L., Liu, C., Freeman, W.T.: Maskgit: Masked generative image transformer. In: CVPR (2022)

    2022

  7. [7]

    In: CVPR

    Chang, H., Zhang, H., Jiang, L., Liu, C., Freeman, W.T.: Maskgit: Masked generative image transformer. In: CVPR. pp. 11305–11315 (2022)

    2022

  8. [8]

    In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G

    Chen, J., Ge, C., Xie, E., Wu, Y., Yao, L., Ren, X., Wang, Z., Luo, P., Lu, H., Li, Z.: Pixart-σ: Weak-to- strong training of diffusion transformer for 4k text-to-image generation. In: Leonardis, A., Ricci, E., Roth, S., Russakovsky, O., Sattler, T., Varol, G. (eds.) Computer Vision – ECCV 2024. pp. 74–91. Springer Nature Switzerland, Cham (2025)

    2024

  9. [9]

    In: Kim, B., Yue, Y., Chaud- huri, S., Fragkiadaki, K., Khan, M., Sun, Y

    Chen, J., YU, J., GE, C., Yao, L., Xie, E., Wang, Z., Kwok, J., Luo, P., Lu, H., Li, Z.: PixArt-α: Fast training of diffusion transformer for photorealistic text-to-image synthesis. In: Kim, B., Yue, Y., Chaud- huri, S., Fragkiadaki, K., Khan, M., Sun, Y. (eds.) International Conference on Learning Representa- tions. vol. 2024, pp. 57611–57640 (2024),http...

    2024

  10. [10]

    In: ICML

    Chen, M., Huang, K., Zhao, T., Wang, M.: Score approximation, estimation and distribution recovery of diffusion models on low-dimensional data. In: ICML. pp. 4672–4712 (2023)

    2023

  11. [11]

    arXiv preprint arXiv:2504.07963 (2025)

    Chen, S., Ge, C., Zhang, S., Sun, P., Luo, P.: Pixelflow: Pixel-space generative models with flow. arXiv preprint arXiv:2504.07963 (2025)

    work page arXiv 2025

  12. [12]

    Scaling Instruction-Finetuned Language Models

    Chung, H.W., Hou, L., Longpre, S., Zoph, B., Tay, Y., Fedus, W., Li, E., Wang, X., Dehghani, M., Brahma, S., Webson, A., Gu, S.S., Dai, Z., Suzgun, M., Chen, X., Chowdhery, A., Narang, S., Mishra, G., Yu, A., Zhao, V., Huang, Y., Dai, A., Yu, H., Petrov, S., Chi, E.H., Dean, J., Devlin, J., Roberts, A., Zhou, D., Le, Q.V., Wei, J.: Scaling instruction-fin...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2210.11416 2022

  13. [13]

    In: NeurIPS (2025)

    Cui, H., Pehlevan, C., Lu, Y.M.: A solvable model of learning generative diffusion: theory and insights. In: NeurIPS (2025)

    2025

  14. [14]

    In: NeurIPS

    De Bortoli, V., Mathieu, E., Hutchinson, M., Thornton, J., Teh, Y.W., Doucet, A.: Riemannian score-based generative modelling. In: NeurIPS. pp. 2406–2422 (2022)

    2022

  15. [15]

    Degeorge, L., Ghosh, A., Dufour, N., Picard, D., Kalogeiton, V.: How far can we go with imagenet for text- to-image generation? arXiv (2025)

    2025

  16. [16]

    In: CVPR

    Deng,J.,Dong,W.,Socher,R.,Li,L.J.,Li,K.,Fei-Fei,L.:Imagenet:Alarge-scalehierarchicalimagedatabase. In: CVPR. pp. 248–255 (2009)

    2009

  17. [17]

    In: NeurIPS

    Dhariwal, P., Nichol, A.: Diffusion models beat gans on image synthesis. In: NeurIPS. pp. 8780–8794 (2021)

    2021

  18. [18]

    In: Proceedings of the 41st International Conference on Machine Learning

    Esser, P., Kulal, S., Blattmann, A., Entezari, R., Müller, J., Saini, H., Levi, Y., Lorenz, D., Sauer, A., Boesel, F., Podell, D., Dockhorn, T., English, Z., Rombach, R.: Scaling rectified flow transformers for high-resolution image synthesis. In: Proceedings of the 41st International Conference on Machine Learning. ICML’24 (2024)

    2024

  19. [19]

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

    Esser, P., Rombach, R., Ommer, B.: Taming transformers for high-resolution image synthesis. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 12873–12883 (2021)

    2021

  20. [20]

    In: The Thirty-ninth Annual Conference on NeurIPS (2025)

    Geng, Z., Deng, M., Bai, X., Kolter, J.Z., He, K.: Mean flows for one-step generative modeling. In: The Thirty-ninth Annual Conference on NeurIPS (2025)

    2025

  21. [22]

    Improved Mean Flows: On the Challenges of Fastforward Generative Models

    Geng, Z., Lu, Y., Wu, Z., Shechtman, E., Kolter, J.Z., He, K.: Improved mean flows: On the challenges of fastforward generative models. arXiv preprint arXiv:2512.02012 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  22. [23]

    In: Proceedings of the 37th International Conference on Neural Information Processing Systems

    Ghosh, D., Hajishirzi, H., Schmidt, L.: Geneval: an object-focused framework for evaluating text-to-image alignment. In: Proceedings of the 37th International Conference on Neural Information Processing Systems. NIPS ’23, Curran Associates Inc., Red Hook, NY, USA (2023)

    2023

  23. [24]

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial networks. Commun. ACM63(11), 139–144 (2020)

    2020

  24. [25]

    In: NeurIPS

    Ho, J., Jain, A., Abbeel, P.: Denoising diffusion probabilistic models. In: NeurIPS. pp. 6840–6851 (2020)

    2020

  25. [26]

    Ho,J.,Salimans,T.:Classifier-freediffusionguidance.In:NeurIPS2021WorkshoponDeepGenerativeModels and Downstream Applications (2021) 20 Duoduo Xue, Zhiyu Zhu, Junhui Hou

    2021

  26. [27]

    In: ICML

    Hoogeboom, E., Heek, J., Salimans, T.: simple diffusion: End-to-end diffusion for high resolution images. In: ICML. pp. 13213–13232 (2023)

    2023

  27. [28]

    In: ICLR (2026)

    Huang, Y., Wang, S.H., Bertozzi, A.L., Wang, B.: RMFlow: Refined mean flow by a noise-injection step for multimodal generation. In: ICLR (2026)

    2026

  28. [29]

    In: NeurIPS (2025)

    Jo, J., Hwang, S.J.: Continuous diffusion model for language modeling. In: NeurIPS (2025)

    2025

  29. [30]

    In: CVPR

    Kang, M., Zhu, J.Y., Zhang, R., Park, J., Shechtman, E., Paris, S., Park, T.: Scaling up gans for text-to-image synthesis. In: CVPR. pp. 10124–10134 (2023)

    2023

  30. [31]

    In: NeurIPS

    Karras, T., Aittala, M., Aila, T., Laine, S.: Elucidating the design space of diffusion-based generative models. In: NeurIPS. pp. 26565–26577 (2022)

    2022

  31. [32]

    In: NeurIPS

    Karras, T., Aittala, M., Kynkäänniemi, T., Lehtinen, J., Aila, T., Laine, S.: Guiding a diffusion model with a bad version of itself. In: NeurIPS. pp. 52996–53021 (2024)

    2024

  32. [33]

    In: CVPR

    Karras, T., Laine, S., Aila, T.: A style-based generator architecture for generative adversarial networks. In: CVPR. pp. 4396–4405 (2019)

    2019

  33. [34]

    IEEE Transactions on Pattern Analysis and Machine Intelligence43(11), 3964–3979 (2021)

    Kobyzev, I., Prince, S.J., Brubaker, M.A.: Normalizing flows: An introduction and review of current methods. IEEE Transactions on Pattern Analysis and Machine Intelligence43(11), 3964–3979 (2021)

    2021

  34. [35]

    Learning on the manifold: Unlocking standard diffusion transformers with representation encoders.arXiv preprint arXiv:2602.10099, 2026

    Kumar, A., Patel, V.M.: Learning on the manifold: Unlocking standard diffusion transformers with represen- tation encoders. arXiv preprint arXiv:2602.10099 (2026)

    work page arXiv 2026

  35. [36]

    In: NeurIPS (2024)

    Kynkäänniemi, T., Aittala, M., Karras, T., Laine, S., Aila, T., Lehtinen, J.: Applying guidance in a limited interval improves sample and distribution quality in diffusion models. In: NeurIPS (2024)

    2024

  36. [37]

    Labs, B.F., Batifol, S., Blattmann, A., Boesel, F., Consul, S., Diagne, C., Dockhorn, T., English, J., English, Z., Esser, P., Kulal, S., Lacey, K., Levi, Y., Li, C., Lorenz, D., Müller, J., Podell, D., Rombach, R., Saini, H., Sauer, A., Smith, L.: Flux.1 kontext: Flow matching for in-context image generation and editing in latent space (2025),https://arx...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [38]

    In: CVPR

    Lee, D., Kim, C., Kim, S., Cho, M., Han, W.S.: Autoregressive image generation using residual quantization. In: CVPR. pp. 11523–11532 (2022)

    2022

  38. [39]

    arXiv preprint arXiv:2504.10483 (2025)

    Leng, X., Singh, J., Hou, Y., Xing, Z., Xie, S., Zheng, L.: Repa-e: Unlocking vae for end-to-end tuning with latent diffusion transformers. arXiv preprint arXiv:2504.10483 (2025)

    work page arXiv 2025

  39. [40]

    In: NeurIPS

    Li, T., Tian, Y., Li, H., Deng, M., He, K.: Autoregressive image generation without vector quantization. In: NeurIPS. vol. 37, pp. 56424–56445 (2024)

    2024

  40. [41]

    arXiv preprint arXiv:2505.21473 (2025)

    Liu, Y., Qu, L., Zhang, H., Wang, X., Jiang, Y., Gao, Y., Ye, H., Li, X., Wang, S., Du, D.K., et al.: Detailflow: 1d coarse-to-fine autoregressive image generation via next-detail prediction. arXiv preprint arXiv:2505.21473 (2025)

    work page arXiv 2025

  41. [42]

    Transactions on Machine Learning Research (2024)

    Loaiza-Ganem, G., Ross, B.L., Hosseinzadeh, R., Caterini, A.L., Cresswell, J.C.: Deep generative models through the lens of the manifold hypothesis: A survey and new connections. Transactions on Machine Learning Research (2024)

    2024

  42. [43]

    In: ICML

    Lou, A., Meng, C., Ermon, S.: Discrete diffusion modeling by estimating the ratios of the data distribution. In: ICML. pp. 32819–32848 (2024)

    2024

  43. [44]

    In: ECCV

    Ma, N., Goldstein, M., Albergo, M.S., Boffi, N.M., Vanden-Eijnden, E., Xie, S.: Sit: Exploring flow and diffusion-based generative models with scalable interpolant transformers. In: ECCV. pp. 23–40 (2024)

    2024

  44. [45]

    In: NeurIPS (2025)

    Ma, X., Zhao, F., Ling, P., Qiu, H., Wei, Z., Yu, H., Huang, J., Zeng, Z., Ma, L.: Towards better & faster autoregressive image generation: From the perspective of entropy. In: NeurIPS (2025)

    2025

  45. [46]

    DeCo: Frequency-Decoupled Pixel Diffusion for End-to-End Image Generation

    Ma, Z., Wei, L., Wang, S., Zhang, S., Tian, Q.: Deco: Frequency-decoupled pixel diffusion for end-to-end image generation. arXiv preprint arXiv:2511.19365 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  46. [47]

    In: ICML

    Nichol, A.Q., Dhariwal, P.: Improved denoising diffusion probabilistic models. In: ICML. pp. 8162–8171 (2021)

    2021

  47. [48]

    In: NeurIPS

    van den Oord, A., Vinyals, O., Kavukcuoglu, K.: Neural discrete representation learning. In: NeurIPS. pp. 6309–6318 (2017)

    2017

  48. [49]

    In: ICCV

    Peebles, W., Xie, S.: Scalable diffusion models with transformers. In: ICCV. pp. 4195–4205 (2023)

    2023

  49. [50]

    In: Kim, B., Yue, Y., Chaud- huri, S., Fragkiadaki, K., Khan, M., Sun, Y

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., Müller, J., Penna, J., Rombach, R.: Sdxl: Improving latent diffusion models for high-resolution image synthesis. In: Kim, B., Yue, Y., Chaud- huri, S., Fragkiadaki, K., Khan, M., Sun, Y. (eds.) International Conference on Learning Representa- tions. vol. 2024, pp. 1862–1874 (2024),https://pr...

    2024

  50. [51]

    In: ICLR (2021)

    Pope, P., Zhu, C., Abdelkader, A., Goldblum, M., Goldstein, T.: The intrinsic dimension of images and its impact on learning. In: ICLR (2021)

    2021

  51. [52]

    In: CVPR

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: CVPR. pp. 10684–10695 (2022)

    2022

  52. [53]

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

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., Ommer, B.: High-resolution image synthesis with latent diffusion models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp. 10684–10695 (June 2022)

    2022

  53. [54]

    In: NeurIPS

    Sahoo, S., Arriola, M., Schiff, Y., Gokaslan, A., Marroquin, E., Chiu, J., Rush, A., Kuleshov, V.: Simple and effective masked diffusion language models. In: NeurIPS. pp. 130136–130184 (2024)

    2024

  54. [55]

    In: ICLR (2022) Diffusion Image Generation with Explicit Modeling of Data Manifold Geometry 21

    Salimans, T., Ho, J.: Progressive distillation for fast sampling of diffusion models. In: ICLR (2022) Diffusion Image Generation with Explicit Modeling of Data Manifold Geometry 21

    2022

  55. [56]

    In: ACM SIG- GRAPH 2022 Conference Proceedings (2022)

    Sauer, A., Schwarz, K., Geiger, A.: Stylegan-xl: Scaling stylegan to large diverse datasets. In: ACM SIG- GRAPH 2022 Conference Proceedings (2022)

    2022

  56. [57]

    In: ICCV

    Shi, F., Luo, Z., Ge, Y., Yang, Y., Shan, Y., Wang, L.: Scalable image tokenization with index backpropagation quantization. In: ICCV. pp. 16037–16046 (2025)

    2025

  57. [58]

    In: NeurIPS (2024)

    Shi, J., Han, K., Wang, Z., Doucet, A., Titsias, M.K.: Simplified and generalized masked diffusion for discrete data. In: NeurIPS (2024)

    2024

  58. [59]

    In: ICLR (2021)

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. In: ICLR (2021)

    2021

  59. [60]

    In: ICML

    Song, Y., Dhariwal, P., Chen, M., Sutskever, I.: Consistency models. In: ICML. pp. 32211–32252 (2023)

    2023

  60. [61]

    In: NeurIPS

    Song, Y., Ermon, S.: Generative modeling by estimating gradients of the data distribution. In: NeurIPS. p. 11895–11907 (2019)

    2019

  61. [62]

    Autoregressive Model Beats Diffusion: Llama for Scalable Image Generation

    Sun, P., Jiang, Y., Chen, S., Zhang, S., Peng, B., Luo, P., Yuan, Z.: Autoregressive model beats diffusion: Llama for scalable image generation. arXiv preprint arXiv:2406.06525 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  62. [63]

    In: Proceedings of The 27th International Conference on Artificial Intelligence and Statistics

    Tang, R., Yang, Y.: Adaptivity of diffusion models to manifold structures. In: Proceedings of The 27th International Conference on Artificial Intelligence and Statistics. pp. 1648–1656 (2024)

    2024

  63. [64]

    In: NeurIPS

    Tian, K., Jiang, Y., Yuan, Z., Peng, B., Wang, L.: Visual autoregressive modeling: scalable image generation via next-scale prediction. In: NeurIPS. pp. 84839–84865 (2024)

    2024

  64. [65]

    In: ICLR 2025 Workshop on Deep Generative Model in Machine Learning: Theory, Principle and Efficacy (2025)

    Wang, P., Zhang, H., Zhang, Z., Chen, S., Ma, Y., Qu, Q.: DIFFUSION MODELS LEARN LOW- DIMENSIONAL DISTRIBUTIONS VIA SUBSPACE CLUSTERING. In: ICLR 2025 Workshop on Deep Generative Model in Machine Learning: Theory, Principle and Efficacy (2025)

    2025

  65. [66]

    Ddt: Decoupled diffusion transformer

    Wang, S., Tian, Z., Huang, W., Wang, L.: Ddt: Decoupled diffusion transformer. arXiv preprint arXiv:2504.05741 (2025)

    work page arXiv 2025

  66. [67]

    In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R

    Xie,E.,Chen,J.,Chen,J.,Cai,H.,Tang,H.,Lin,Y.,Zhang,Z.,Li,M.,Zhu,L.,Lu,Y.,Han,S.:Sana:Efficient high-resolution text-to-image synthesis with linear diffusion transformers. In: Yue, Y., Garg, A., Peng, N., Sha, F., Yu, R. (eds.) International Conference on Learning Representations. vol. 2025, pp. 20383–20407 (2025), https://proceedings.iclr.cc/paper_files/p...

    2025

  67. [68]

    Transactions on Machine Learning Research (2025)

    Xiong, J., Liu, G., Huang, L., Wu, C., Wu, T., Mu, Y., Yao, Y., Shen, H., Wan, Z., Huang, J., et al.: Autoregressive models in vision: A survey. Transactions on Machine Learning Research (2025)

    2025

  68. [69]

    Representation Fr\'echet Loss for Visual Generation

    Yang, J., Geng, Z., Ju, X., Tian, Y., Wang, Y.: Representation fréchet loss for visual generation. arXiv:2604.28190 (2026),https://arxiv.org/abs/2604.28190

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  69. [70]

    In: ICML (2025)

    You, Z., Ou, J., Zhang, X., Hu, J., ZHOU, J., Li, C.: Effective and efficient masked image generation models. In: ICML (2025)

    2025

  70. [71]

    In: ICLR (2022)

    Yu, J., Li, X., Koh, J.Y., Zhang, H., Pang, R., Qin, J., Ku, A., Xu, Y., Baldridge, J., Wu, Y.: Vector-quantized image modeling with improved VQGAN. In: ICLR (2022)

    2022

  71. [72]

    In: ICLR (2024)

    Yu, L., Lezama, J., Gundavarapu, N.B., Versari, L., Sohn, K., Minnen, D., Cheng, Y., Gupta, A., Gu, X., Hauptmann, A.G., Gong, B., Yang, M.H., Essa, I., Ross, D.A., Jiang, L.: Language model beats diffusion - tokenizer is key to visual generation. In: ICLR (2024)

    2024

  72. [73]

    In: ICLR (2025)

    Yu, S., Kwak, S., Jang, H., Jeong, J., Huang, J., Shin, J., Xie, S.: Representation alignment for generation: Training diffusion transformers is easier than you think. In: ICLR (2025)

    2025

  73. [74]

    In: CVPR

    Zhang, R., Isola, P., Efros, A.A., Shechtman, E., Wang, O.: The unreasonable effectiveness of deep features as a perceptual metric. In: CVPR. pp. 586–595 (2018)

    2018

  74. [75]

    In: NeurIPS (2025)

    Zheng, A., Wen, X., Zhang, X., Ma, C., Wang, T., YU, G., Zhang, X., QI, X.: Vision foundation models as effective visual tokenizers for autoregressive generation. In: NeurIPS (2025)

    2025

  75. [76]

    In: ICLR (2026)

    Zheng, B., Ma, N., Tong, S., Xie, S.: Diffusion transformers with representation autoencoders. In: ICLR (2026)

    2026

  76. [77]

    In: NeurIPS

    Zhu, L., Wei, F., Lu, Y., Chen, D.: Scaling the codebook size of vqgan to 100,000 with a utilization rate of 99%. In: NeurIPS. pp. 12612–12635 (2024)

    2024

  77. [78]

    Feature" level inherently retains all continuous architectural benefits (including our dual-branch high-frequency embeddings), the strict superiority of “Logits

    Zhuo, L., Du, R., Xiao, H., Li, Y., Liu, D., Huang, R., Liu, W., Zhu, X., Wang, F.Y., Ma, Z., Luo, X., Wang, Z., Zhang, K., Zhao, L., Liu, S., Yue, X., Ouyang, W., Qiao, Y., Li, H., Gao, P.: Lumina-next : Making lumina-t2x stronger and faster with next-dit. In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (2024),https://ope...

    2024

  78. [79]

    Diffusion & Noise Settings Noise Scale Factorc1 0.6 0.8 Signal Scale Factorc2 1.0 1.0 Training Timestep Range (t)t∈[0.2,0.95]t∈[0.2,0.95]

  79. [80]

    Embedding & Vocabulary Vocabulary SizeV16,384 8192 Embedding DimensionL16 8 Embedding Subspace 4 2

  80. [81]

    Network Architecture Total Parameters(M) 130.48 35.21 Number of Blocks 14 14 Hidden Size 768 384 Attention Heads 12 6 Condition Embedding Dimension 128 128

Showing first 80 references.