Pith. sign in

REVIEW 2 major objections 4 minor 2 cited by

D-AR: Diffusion via Autoregressive Models

T0 review · 2 major / 4 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 D-AR recasts image diffusion as a vanilla autoregressive next-token prediction problem, and shows that a 775M Llama-style transformer generates ImageNet 256×256 images at 2.09 FID with 256 discrete tokens while natively supporting…

desk verdict Diffusion-ordered discrete tokens let a vanilla Llama produce competitive ImageNet FID; the architecture is real, but preview consistency needs a quantitative metric before I'd call the headline property proven. read the letter →

arxiv 2505.23660 v1 pith:KG7A5XYA submitted 2025-05-29 cs.CV

Ziteng Gao , Mike Zheng Shou This is my paper
classification cs.CV
keywords diffusionmodelsautoregressiveimagegenerationsequentialtokenizernext-tokenpredictionflowmatchingvisualtokenizationNetclass-conditionalKVcache
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

Diffusion via Autoregressive models (D-AR) claims that the image diffusion process can be recast as a standard autoregressive next-token prediction problem, with no changes to the causal mask, loss, or inference loop of a vanilla LLM-style transformer. The paper builds a sequential diffusion tokenizer that maps an image to 256 discrete tokens, with later tokens feeding later denoising steps of a pixel-space flow-matching decoder, so autoregressive token generation directly mirrors diffusion denoising. On ImageNet 256×256, D-AR-XL (775M parameters) achieves 2.09 FID, outperforming vanilla autoregressive baselines and matching much larger models, while gaining native KV-cache inference, consistent partial-token previews, and zero-shot layout condition from prefix tokens. The significance is architectural: it suggests LLM stacks need no vision-specific modifications to do diffusion-quality generation.

What carries the argument

The sequential diffusion tokenizer: a transformer encoder with learnable query tokens and vector quantization produces 256 ordered discrete tokens, and a 185M-parameter diffusion transformer decoder takes, at each flow-matching timestep $t$, only the token group $g_{\lceil t' K \rceil}$ selected by the shifted condition schedule $c(t)=g_{\lceil t' K \rceil}$ with $t'=t/(t+(1/\beta)(1-t))$. This makes diffusion look like a Markov chain of conditioning groups, so an LLM generating tokens one at a time is literally supplying the next denoising condition. The reverse schedule $t_i = (i/K)/(i/K+\beta(1-i/K))$ fixes 8 denoising steps, each using one group of 32 tokens.

What would settle it

Swap two token groups at decode time (for example, feed group 2's tokens as the condition for the first denoising step) and measure FID and layout similarity against the original schedule; if quality and coarse structure do not collapse, the claimed coarse-to-fine ordering is not actually determining the output, and the central equivalence claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that a discrete token sequence can serve as a compressed control program for a diffusion process: instead of sampling noise and iterating a denoiser, one samples tokens left to right, and each group of tokens conditions the next denoising step in pixel space. Because the condition schedule $c(t)$ reads token groups in order as the flow-matching time $t$ goes from 0 to 1, early tokens encode coarse layout and global structure, while later tokens add fine detail, producing a natural linear ordering for autoregressive generation. Training uses standard cross-entropy on discrete tokens plus a flow-matching tokenizer loss with perceptual and representation-alignment terms; decoding uses a fixed 8-step reverse schedule with each of the $K=8$ groups consumed exactly once. The paper shows that this yields state-of-the-art results among vanilla (unmodified) autoregressive image models on ImageNet, and that partial token sequences can be decoded immediately to produce consistent coarse-to-fine previews.

Load-bearing premise

The load-bearing premise is that each diffusion denoising step needs only its own group of $N/K$ tokens as conditioning, consumed exactly once in a fixed order through the schedule $c(t)$; if later tokens carry information required at earlier denoising steps, the equivalence between autoregressive token generation and the diffusion procedure breaks.

Editorial extensions

If this is right

  • LLM serving infrastructure such as KV-cache acceleration applies unchanged to high-quality image generation via D-AR, since the image model is a vanilla decoder-only transformer.
  • Streaming generation gives consistent previews at any partial token count with near-zero extra cost, because the pixel-space diffusion decoder can jump-estimate the target image from whatever condition tokens are available.
  • Layout-controlled synthesis is achievable zero-shot by fixing a few prefix tokens as conditions, with no finetuning, and the control strengthens as more prefix tokens are provided.
  • D-AR-XL reaches 2.09 FID with 775M parameters, outperforming LlamaGen-XXL at 1.4B and matching IBQ-XXL at 2.1B within the vanilla autoregressive category.
  • The sequential diffusion tokenizer reconstructs ImageNet validation images at rFID 1.58 with 256 tokens and a 16384-entry codebook, beating LlamaGen's tokenizer at the same budget (2.19 rFID), and degrades less when the codebook is shrunk.

Reading between the lines

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

  • If the schedule sufficiency holds beyond the tested regime, the same conditioning-group trick could linearize other coarse-to-fine generative processes (video, audio, 3D) into vanilla autoregression, but the paper explicitly leaves native text-to-image and parameter scales above 1B untested, so that breadth is extrapolation.
  • The fixed $K=8$ reverse schedule is a design choice rather than a necessity; an adaptive schedule where later groups re-condition earlier timesteps could test whether the Markovian claim survives without the fixed-step implementation.
  • If the coarse-to-fine ordering is genuinely load-bearing, token count may scale with the number of diffusion steps rather than spatial resolution, so high-resolution generation might stay token-cheap and fit within LLM context windows.
  • Because D-AR outputs discrete tokens in a standard causal format, it should integrate directly into multimodal LLM pipelines, but the paper does not benchmark wall-clock speedups, so the practical speed advantage remains an invitation rather than a measured result.
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

2 major / 4 minor

Summary. The paper proposes D-AR, a framework that bridges diffusion and autoregressive visual generation by introducing a 'sequential diffusion tokenizer.' This tokenizer maps an image into a 1D sequence of 256 discrete tokens and is trained so that successive groups of tokens condition successive timesteps of a flow-matching diffusion decoder on raw pixels. A standard decoder-only Llama-style transformer then performs vanilla next-token prediction on these tokens, and the resulting token generation can be incrementally decoded into diffusion denoising steps. The paper reports a reconstruction rFID of 1.52 for the tokenizer and an ImageNet 256x256 class-conditional generation FID of 2.09 for D-AR-XL (775M AR parameters), along with claimed native support for KV-cache inference, consistent streaming previews from partial tokens, and zero-shot layout-controlled synthesis.

Significance. If the claims are substantiated, D-AR offers a conceptually clean way to reuse standard LLM autoregressive infrastructure for image generation while retaining diffusion-style coarse-to-fine decoding. The experiments are reasonably extensive for a preprint: the paper reports tokenizer ablations over the conditioning group count K and the timeshift parameter beta, compares against several strong baselines, and is transparent about the tokenizer parameter count in a table note. The reported 2.09 FID with a 775M AR backbone is competitive among vanilla autoregressive methods. The main value is the sequential diffusion tokenizer idea, which is novel relative to prior tokenizers with diffusion decoders. The primary weakness is that two of the advertised properties, consistent streaming previews and zero-shot layout control, are supported only by qualitative examples and distribution-level FID numbers, not by a direct consistency evaluation.

major comments (2)
  1. [§5.1, Table 3] The headline parameter count of 775M for D-AR-XL counts only the autoregressive backbone and excludes the 300M sequential diffusion tokenizer. While the Table 3 note discloses this, the abstract, introduction, and conclusion repeat the 775M figure without the caveat, and the surrounding text compares D-AR-XL to IBQ-XXL 2.1B as if the full systems were comparable. Please report total system parameters (1.075B for D-AR-XL) in the headline statements and revisit the parameter-region comparisons, or clearly state that all compared methods exclude tokenizer parameters and provide the tokenizer budget for each.
  2. [§3.1, Eq. (5); §3.3, Eq. (8); §5.1, Table 5] The claimed property of 'consistent previews' rests on the sufficiency of the fixed condition schedule c(t), namely that each disjoint group of tokens consumed in order provides the conditioning needed at its diffusion step. The quantitative evidence in Table 5, however, consists of FID, IS, precision, and recall of the partial-token jump estimates, which are distribution-level metrics and do not establish whether a preview at, say, 64 tokens is the same image as the final sample at 256 tokens. Please add a paired consistency metric (for instance, LPIPS or segmentation/label agreement between each preview and its own final image, or a same-prefix-multiple-completions study) and report the numbers. Without this, the streaming-preview property is supported only by the qualitative Figures 5 and 11, which is insufficient for a headline claim.
minor comments (4)
  1. [Section 1] The text contains a typo: 'Dffusion via Autoregressive models' should read 'Diffusion via Autoregressive models.'
  2. [Appendix III] The phrase 'coodebook utilization' should be corrected to 'codebook utilization.'
  3. [Eq. (5) and Eq. (8)] The statement that Eq. (8) is the 'reversed form' of Eq. (5) is ambiguous: Eq. (8) is not the algebraic inverse of Eq. (5). Please clarify the exact relationship between these two schedules.
  4. [§5.1, Zero-shot layout-controlled synthesis] The zero-shot layout-controlled synthesis property is currently demonstrated only with qualitative examples (Figure 6 and Figure 12). If this is retained as a headline property, a quantitative layout-fidelity metric or a comparison with a simple conditioning baseline would strengthen the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: D-AR's construction is architectural, its benchmarks are external, and self-citations are not load-bearing.

full rationale

The paper derives no numerical prediction from fitted constants. The sequential diffusion tokenizer is trained with Eq. 2/7 to associate token groups with diffusion timesteps; the AR model is trained with standard cross-entropy (Eq. 6); system FID (Table 3) is measured against an external ImageNet benchmark. The 'consistent previews' property follows from the jump-estimate formula x1hat=(1-t)v_t+x_t applied to the same decoder used in sampling, so the preview is the model's own estimate at partial tokens; whether it agrees with the final image is an empirical property shown in Fig. 5, not a fitted value renamed as a prediction. The conditioning schedule c(t) (Eq. 5) and sampler (Eq. 8) are explicit design choices, ablated in Tables 6-7, and their sufficiency for good generation is an empirical claim, not a circular one. Self-citations (Refs. 12 and 39, sharing author M.Z. Shou) appear only in the related-work survey and are not load-bearing. The layout-control and coarse-to-fine order are built into the tokenizer by design; presenting them as properties of the architecture is a construction claim, not a derivation. The reader-flagged concern that preview consistency lacks a quantitative metric is a validation gap for correctness, not circularity. Under the stated rules, no step reduces by definition or by self-citation to its own inputs.

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

The framework's validity rests on standard flow matching assumptions, the empirical generalization of the conditioning schedule, and the learnability of the diffusion-induced token order by a vanilla causal AR model. The central hyperparameters beta=2, K=8, CFG schedules, and loss weights are hand-selected via ablations rather than derived. No new physical entities are introduced; the sequential diffusion tokenizer is a composition of existing components.

free parameters (5)
  • beta (timeshift control) = 2
    Hand-chosen via ablation in Table 6; controls how many token groups are spent on early denoising steps. Central to the condition schedule in Eq. (5).
  • K (number of conditioning groups / sampling steps) = 8
    Chosen via ablation in Table 7 as a trade-off between reconstruction FID and linearized token structure for AR training.
  • CFG schedule for D-AR-L = 1.1 to 8.0 linear
    Selected per model; from Table 4, FID drops from 7.43 (no CFG) to 2.44 with this schedule.
  • CFG schedule for D-AR-XL = 1.1 to 10.0 linear
    Selected per model; from Table 4, FID drops from 5.11 (no CFG) to 2.09 with this schedule.
  • loss weights lambda1, lambda2 = 0.5 each
    Hand-assigned in Eq. (7) for LPIPS and REPA losses; not ablated.
assumptions (4)
  • standard math Flow matching with velocity target v_t = x1 - x0 and linear interpolant yields valid generative ODE sampling.
    Invoked in Eqs. (2)-(4) and the sampling procedure in Section 4; standard in flow matching literature.
  • domain assumption The diffusion decoder trained with a randomly sampled timestep and only the corresponding group of tokens generalizes to a fixed schedule that consumes each group exactly once.
    The 8-step schedule in Eq. (8) is an approximation; the paper relies on empirical FID rather than a proof that group-wise conditioning is sufficient.
  • domain assumption The token sequence induced by the diffusion schedule has a learnable coarse-to-fine conditional structure for vanilla next-token prediction.
    The AR model's success assumes p(zi|z<i) is learnable; this is an empirical premise validated only by the final FID.
  • domain assumption Causal masking of query tokens in the encoder and causal decoder layers preserves enough information for generation.
    Design choice in Section 4; no analysis of information loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of D-AR: Diffusion via Autoregressive Models." pith.science (2026). https://pith.science/paper/KG7A5XYA

@misc{pith2026250523660,
  author       = {Pith},
  title        = {Pith review of: D-AR: Diffusion via Autoregressive Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KG7A5XYA}},
  note         = {Machine review of arXiv:2505.23660}
}
read the original abstract

This paper presents Diffusion via Autoregressive models (D-AR), a new paradigm recasting the image diffusion process as a vanilla autoregressive procedure in the standard next-token-prediction fashion. We start by designing the tokenizer that converts images into sequences of discrete tokens, where tokens in different positions can be decoded into different diffusion denoising steps in the pixel space. Thanks to the diffusion properties, these tokens naturally follow a coarse-to-fine order, which directly lends itself to autoregressive modeling. Therefore, we apply standard next-token prediction on these tokens, without modifying any underlying designs (either causal masks or training/inference strategies), and such sequential autoregressive token generation directly mirrors the diffusion procedure in image space. That is, once the autoregressive model generates an increment of tokens, we can directly decode these tokens into the corresponding diffusion denoising step in the streaming manner. Our pipeline naturally reveals several intriguing properties, for example, it supports consistent previews when generating only a subset of tokens and enables zero-shot layout-controlled synthesis. On the standard ImageNet benchmark, our method achieves 2.09 FID using a 775M Llama backbone with 256 discrete tokens. We hope our work can inspire future research on unified autoregressive architectures of visual synthesis, especially with large language models. Code and models will be available at https://github.com/showlab/D-AR

Figures

Figures reproduced from arXiv: 2505.23660 by the authors.

Figure 1
Figure 1. Diffusion via autoregressive modeling (D-AR) framework for visual generation. As the autoregressive transformer generates tokens, D-AR can simultaneously perform corresponding diffusion steps via token conditioning and jump-estimate target samples as rough previews effortlessly. Abstract This paper presents Diffusion via Autoregressive models (D-AR), a new paradigm recasting the image diffusion process as a vanilla … view at source ↗
Figure 2
Figure 2. Uncurated generated samples from D-AR-XL with 256 × 256 resolutions (CFG=4.0). 1 Introduction Autoregressive models, exemplified by large language models (LLMs) [1, 2, 3], have emerged as the foundation of modern NLP, achieving state-of-the-art performance with the simple next token prediction paradigm. With their widespread adoption, the autoregressive next token prediction paradigm has established itself as the de… view at source ↗
Figure 3
Figure 3. Different paradigms incorporating diffusion and autoregressive models for vision generation. (a) uses patch-level diffusion during every single autoregressive step to tackle continuous outputs [27, 28, 29]; (b) DART [30] denoises a full image per every autoregressive step by AR transformer, together inputted with history denoised images; (c) use a single set of continuous outputs to different diffusion steps and req… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Sequential diffusion tokenizer structure. When training the tokenizer, the pixel diffusion transformer in the tokenizer decoder calculates the velocity loss with the selected group of tokens, c(t), as conditioning tokens. semantics of images and extract such semantics …
Figure 5
Figure 5. Figure 5: Consistent previews as generation trajectories for every increment of 32 tokens (a group). Note that these previews can be generated in a streaming manner with AR tokens partially generated. reference 8 prefix tokens 16 prefix tokens 12 prefix tokens 32 prefix tokens “…
Figure 6
Figure 6. Figure 6 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Vanilla AR comparison for ImageNet generation. In this paper, we present Diffusion via Autoregressive models (D-AR), a framework to bridge the pixel diffusion and autoregressive modeling for visual generation. With the linearized sequence of discrete tokens by the pres…
Figure 8
Figure 8. Figure 8: Uncurated generated samples by D-AR-XL with random labels and CFG=4.0. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Uncurated generated samples by D-AR-XL with random labels and CFG=4.0 (cont’d). 13 [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Reconstruction results with samples from the ImageNet validation set. Each pair of rows shows: first row — input; second row — reconstruction. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_10.png]
Figure 11
Figure 11. Figure 11: Generation trajectory and previews at each diffusion sampling step by D-AR-L. reference 8 prefix tokens 16 prefix tokens 12 prefix tokens 32 prefix tokens “giant panda” “golden retriever” “otter” [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Zero-shot layout-controlled synthesis. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Learning to Tessellate: Point Cloud Generation via Recursive Spectral Partitioning

    cs.CV 2026-08 conditional novelty 6.0 of 10

    PointRSP recursively partitions a point cloud's KNN graph into a non-balanced binary tree and autoregressively generates cluster structure and residual features, improving 1-NNA on ShapeNet.

  2. EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models

    cs.CV 2026-03 conditional novelty 6.0 of 10

    Iterative latent thought refinement plus terminal text grounding lets diffusion models solve multi-step visual reasoning tasks at 92.1% average accuracy, beating DiffThinker by 8.3 points.

Reference graph

Works this paper leans on

78 extracted references · 52 canonical work pages · cited by 2 Pith papers

  1. [1]

    Gpt-4 technical report, 2024

    GPT-4 Team. Gpt-4 technical report, 2024. 2, 3

    work page 2024

  2. [2]

    Llama: Open and efficient foundation language models, 2023

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language models, 2023. 2, 3, 4, 5

    work page 2023

  3. [3]

    The llama 3 herd of models, 2024

    Llama 3 Team. The llama 3 herd of models, 2024. 2, 3

    work page 2024

  4. [4]

    Megatron-lm: Training multi-billion parameter language models using model parallelism, 2020

    Mohammad Shoeybi, Mostofa Patwary, Raul Puri, Patrick LeGresley, Jared Casper, and Bryan Catanzaro. Megatron-lm: Training multi-billion parameter language models using model parallelism, 2020. 2

    work page 2020

  5. [5]

    Efficient memory management for large language model serving with pagedattention

    Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, and Ion Stoica. Efficient memory management for large language model serving with pagedattention. InSOSP, pages 611–626. ACM, 2023. 2

    work page 2023

  6. [6]

    Gonzalez, Clark W

    Lianmin Zheng, Liangsheng Yin, Zhiqiang Xie, Chuyue Sun, Jeff Huang, Cody Hao Yu, Shiyi Cao, Christos Kozyrakis, Ion Stoica, Joseph E. Gonzalez, Clark W. Barrett, and Ying Sheng. Sglang: Efficient execution of structured language model programs. InNeurIPS, 2024. 2

    work page 2024

  7. [7]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis. InCVPR, pages 12873–12883. Computer Vision Foundation / IEEE, 2021. 2, 3

    work page 2021

  8. [8]

    Autore- gressive model beats diffusion: Llama for scalable image generation.arXiv, abs/2406.06525, 2024

    Peize Sun, Yi Jiang, Shoufa Chen, Shilong Zhang, Bingyue Peng, Ping Luo, and Zehuan Yuan. Autore- gressive model beats diffusion: Llama for scalable image generation.arXiv, abs/2406.06525, 2024. 2, 3, 4, 5, 6, 7, 8, 10

    arXiv 2024

Show all 78 references
  1. [9]

    Autoregressive image generation using residual quantization

    Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. Autoregressive image generation using residual quantization. InCVPR, pages 11513–11522. IEEE, 2022. 2, 3, 7

    2022

  2. [10]

    Chameleon: Mixed-modal early-fusion foundation models.arXiv, abs/2405.09818,

    Chameleon Team. Chameleon: Mixed-modal early-fusion foundation models.arXiv, abs/2405.09818,

    arXiv

  3. [11]

    Transfusion: Predict the next token and diffuse images with one multi-modal model

    Chunting Zhou, Lili Yu, Arun Babu, Kushal Tirumala, Michihiro Yasunaga, Leonid Shamis, Jacob Kahn, Xuezhe Ma, Luke Zettlemoyer, and Omer Levy. Transfusion: Predict the next token and diffuse images with one multi-modal model. InICLR. OpenReview.net, 2025. 2, 3

    2025

  4. [12]

    Show-o: One single transformer to unify multimodal understanding and generation

    Jinheng Xie, Weijia Mao, Zechen Bai, David Junhao Zhang, Weihao Wang, Kevin Qinghong Lin, Yuchao Gu, Zhijie Chen, Zhenheng Yang, and Mike Zheng Shou. Show-o: One single transformer to unify multimodal understanding and generation. InICLR. OpenReview.net, 2025. 2

    2025

  5. [13]

    Janusflow: Harmonizing autoregression and rectified flow for unified multimodal understanding and generation.arXiv, abs/2411.07975, 2024

    Yiyang Ma, Xingchao Liu, Xiaokang Chen, Wen Liu, Chengyue Wu, Zhiyu Wu, Zizheng Pan, Zhenda Xie, Haowei Zhang, Xingkai Yu, Liang Zhao, Yisong Wang, Jiaying Liu, and Chong Ruan. Janusflow: Harmonizing autoregression and rectified flow for unified multimodal understanding and ge...

    2024 arXiv

  6. [14]

    Visual autoregressive modeling: Scalable image generation via next-scale prediction

    Keyu Tian, Yi Jiang, Zehuan Yuan, Bingyue Peng, and Liwei Wang. Visual autoregressive modeling: Scalable image generation via next-scale prediction. InNeurIPS, 2024. 2, 3, 4, 8

    2024

  7. [15]

    Freeman, and Yu-Xiong Wang

    Ziqi Pang, Tianyuan Zhang, Fujun Luan, Yunze Man, Hao Tan, Kai Zhang, William T. Freeman, and Yu-Xiong Wang. Randar: Decoder-only autoregressive visual generation in random orders.arXiv, abs/2412.01827, 2024. 2, 3, 4, 7, 8, 10

    2024 arXiv

  8. [16]

    Randomized autoregressive visual generation.arXiv, abs/2411.00776, 2024

    Qihang Yu, Ju He, Xueqing Deng, Xiaohui Shen, and Liang-Chieh Chen. Randomized autoregressive visual generation.arXiv, abs/2411.00776, 2024. 2, 3, 4, 8, 10

    2024 arXiv

  9. [17]

    Yuille, and Liang-Chieh Chen

    Sucheng Ren, Qihang Yu, Ju He, Xiaohui Shen, Alan L. Yuille, and Liang-Chieh Chen. Beyond next-token: Next-x prediction for autoregressive visual generation.arXiv, abs/2502.20388, 2025. 2, 3, 4

    2025 arXiv

  10. [18]

    Imagefolder: Autoregressive image generation with folded tokens

    Xiang Li, Kai Qiu, Hao Chen, Jason Kuen, Jiuxiang Gu, Bhiksha Raj, and Zhe Lin. Imagefolder: Autoregressive image generation with folded tokens. InICLR, 2025. 2

    2025

  11. [19]

    Denoising diffusion implicit models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models. InICLR. OpenReview.net, 2021. 2, 3

    2021

  12. [20]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. InNeurIPS, 2020. 2, 3 16

    2020

  13. [21]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matthew Le. Flow matching for generative modeling. InICLR. OpenReview.net, 2023. 2, 3, 5

    2023

  14. [22]

    Zero-shot text-to-image generation

    Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea V oss, Alec Radford, Mark Chen, and Ilya Sutskever. Zero-shot text-to-image generation. InICML, volume 139 ofProceedings of Machine Learning Research, pages 8821–8831. PMLR, 2021. 2, 3

    2021

  15. [23]

    Flux.https://github.com/black-forest-labs/flux, 2023

    Black Forest Labs. Flux.https://github.com/black-forest-labs/flux, 2023. 2, 3

    2023

  16. [24]

    SDXL: improving latent diffusion models for high-resolution image synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. SDXL: improving latent diffusion models for high-resolution image synthesis. In ICLR. OpenReview.net, 2024. 2

    2024

  17. [25]

    Generative modelling with inverse heat dissipation

    Severi Rissanen, Markus Heinonen, and Arno Solin. Generative modelling with inverse heat dissipation. InICLR. OpenReview.net, 2023. 2, 5, 9

    2023

  18. [26]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Fei-Fei Li. Imagenet: A large-scale hierarchical image database. InCVPR, 2009. 2, 7

    2009

  19. [27]

    Autoregressive image generation without vector quantization

    Tianhong Li, Yonglong Tian, He Li, Mingyang Deng, and Kaiming He. Autoregressive image generation without vector quantization. InNeurIPS, 2024. 3, 8, 10

    2024

  20. [28]

    Causal diffusion transformers for generative modeling.arXiv, abs/2412.12095, 2024

    Chaorui Deng, Deyao Zhu, Kunchang Li, Shi Guang, and Haoqi Fan. Causal diffusion transformers for generative modeling.arXiv, abs/2412.12095, 2024. 3, 8

    2024 arXiv

  21. [29]

    Diffusion forcing: Next-token prediction meets full-sequence diffusion

    Boyuan Chen, Diego Marti Monso, Yilun Du, Max Simchowitz, Russ Tedrake, and Vincent Sitzmann. Diffusion forcing: Next-token prediction meets full-sequence diffusion. InNeurIPS, 2024. 3

    2024

  22. [30]

    Denoising autoregressive transformers for scalable text-to-image generation

    Jiatao Gu, Yuyang Wang, Yizhe Zhang, Qihang Zhang, Dinghuai Zhang, Navdeep Jaitly, Joshua M Susskind, and Shuangfei Zhai. Denoising autoregressive transformers for scalable text-to-image generation. InThe Thirteenth International Conference on Learning Representations, 2025. 3, 6, 8

    2025

  23. [31]

    Dreamllm: Synergistic multimodal comprehension and creation

    Runpei Dong, Chunrui Han, Yuang Peng, Zekun Qi, Zheng Ge, Jinrong Yang, Liang Zhao, Jianjian Sun, Hongyu Zhou, Haoran Wei, Xiangwen Kong, Xiangyu Zhang, Kaisheng Ma, and Li Yi. Dreamllm: Synergistic multimodal comprehension and creation. InICLR. OpenReview.net, 2024. 3

    2024

  24. [32]

    SEED-X: multimodal models with unified multi-granularity comprehension and generation.arXiv, abs/2404.14396, 2024

    Yuying Ge, Sijie Zhao, Jinguo Zhu, Yixiao Ge, Kun Yi, Lin Song, Chen Li, Xiaohan Ding, and Ying Shan. SEED-X: multimodal models with unified multi-granularity comprehension and generation.arXiv, abs/2404.14396, 2024. 3

    2024 arXiv

  25. [33]

    Next-gpt: Any-to-any multimodal llm

    Shengqiong Wu, Hao Fei, Leigang Qu, Wei Ji, and Tat-Seng Chua. Next-gpt: Any-to-any multimodal llm. InForty-first International Conference on Machine Learning, 2024. 3

    2024

  26. [34]

    Transfer between modalities with metaqueries.arXiv, abs/2504.06256, 2025

    Xichen Pan, Satya Narayan Shukla, Aashu Singh, Zhuokai Zhao, Shlok Kumar Mishra, Jialiang Wang, Zhiyang Xu, Jiuhai Chen, Kunpeng Li, Felix Juefei-Xu, Ji Hou, and Saining Xie. Transfer between modalities with metaqueries.arXiv, abs/2504.06256, 2025. 3

    2025 arXiv

  27. [35]

    Flow straight and fast: Learning to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow. InICLR. OpenReview.net, 2023. 3, 5

    2023

  28. [36]

    Gemini: A family of highly capable multimodal models, 2025

    Gemini Team. Gemini: A family of highly capable multimodal models, 2025. 3

    2025

  29. [37]

    Qwen technical report, 2023

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Si...

    2023

  30. [38]

    Neural discrete representation learning

    Aäron van den Oord, Oriol Vinyals, and Koray Kavukcuoglu. Neural discrete representation learning. In NIPS, pages 6306–6315, 2017. 3, 4, 10

    2017

  31. [39]

    Long-context autoregressive video modeling with next-frame prediction.CoRR, abs/2503.19325, 2025

    Yuchao Gu, Weijia Mao, and Mike Zheng Shou. Long-context autoregressive video modeling with next-frame prediction.CoRR, abs/2503.19325, 2025. 3

    2025 arXiv

  32. [40]

    Vector-quantized image modeling with improved VQGAN

    Jiahui Yu, Xin Li, Jing Yu Koh, Han Zhang, Ruoming Pang, James Qin, Alexander Ku, Yuanzhong Xu, Jason Baldridge, and Yonghui Wu. Vector-quantized image modeling with improved VQGAN. InICLR. OpenReview.net, 2022. 3 17

    2022

  33. [41]

    consistencydecoder.https://github.com/openai/consistencydecoder, 2023

    OpenAI. consistencydecoder.https://github.com/openai/consistencydecoder, 2023. 3

    2023

  34. [42]

    Long Zhao, Sanghyun Woo, Ziyu Wan, Yandong Li, Han Zhang, Boqing Gong, Hartwig Adam, Xuhui Jia, and Ting Liu.ϵ-vae: Denoising as visual decoding.arXiv, abs/2410.04081, 2024. 3, 6

    2024 arXiv

  35. [43]

    HART: efficient visual generation with hybrid autoregressive transformer

    Haotian Tang, Yecheng Wu, Shang Yang, Enze Xie, Junsong Chen, Junyu Chen, Zhuoyang Zhang, Han Cai, Yao Lu, and Song Han. HART: efficient visual generation with hybrid autoregressive transformer. arXiv, abs/2410.10812, 2024. 3

    2024 arXiv

  36. [44]

    Flow to the mode: Mode-seeking diffusion autoencoders for state-of-the-art image tokenization.arXiv, abs/2503.11056, 2025

    Kyle Sargent, Kyle Hsu, Justin Johnson, Li Fei-Fei, and Jiajun Wu. Flow to the mode: Mode-seeking diffusion autoencoders for state-of-the-art image tokenization.arXiv, abs/2503.11056, 2025. 3, 11

    2025

  37. [45]

    Diffusion autoen- coders are scalable image tokenizers.arXiv, abs/2501.18593, 2025

    Yinbo Chen, Rohit Girdhar, Xiaolong Wang, Sai Saketh Rambhatla, and Ishan Misra. Diffusion autoen- coders are scalable image tokenizers.arXiv, abs/2501.18593, 2025. 3, 11

    2025 arXiv

  38. [46]

    principal components

    Xin Wen, Bingchen Zhao, Ismail Elezi, Jiankang Deng, and Xiaojuan Qi. "principal components" enable A new language of images.arXiv, abs/2503.08685, 2025. 4

    2025 arXiv

  39. [47]

    Flextok: Resampling images into 1d token sequences of flexible length.arXiv, abs/2502.13967, 2025

    Roman Bachmann, Jesse Allardice, David Mizrahi, Enrico Fini, Oguzhan Fatih Kar, Elmira Amirloo, Alaaeldin El-Nouby, Amir Zamir, and Afshin Dehghan. Flextok: Resampling images into 1d token sequences of flexible length.arXiv, abs/2502.13967, 2025. 4

    2025 arXiv

  40. [48]

    Kingma and Max Welling

    Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. InICLR, 2014. 4, 6

    2014

  41. [49]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InCVPR, pages 10674–10685. IEEE, 2022. 4, 6

    2022

  42. [50]

    Generative multimodal pretraining with discrete diffusion timestep tokens.arXiv, abs/2504.14666, 2025

    Kaihang Pan, Wang Lin, Zhongqi Yue, Tenglong Ao, Liyu Jia, Wei Zhao, Juncheng Li, Siliang Tang, and Hanwang Zhang. Generative multimodal pretraining with discrete diffusion timestep tokens.arXiv, abs/2504.14666, 2025. 4

    2025 arXiv

  43. [51]

    An image is worth 32 tokens for reconstruction and generation.Advances in Neural Information Processing Systems, 37:128940–128966, 2024

    Qihang Yu, Mark Weber, Xueqing Deng, Xiaohui Shen, Daniel Cremers, and Liang-Chieh Chen. An image is worth 32 tokens for reconstruction and generation.Advances in Neural Information Processing Systems, 37:128940–128966, 2024. 4, 6, 7, 8

    2024

  44. [52]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. InNIPS, pages 5998–6008, 2017. 4

    2017

  45. [53]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InICCV, pages 4172–4182. IEEE, 2023. 5, 6, 8

    2023

  46. [54]

    Albergo, Nicholas M

    Nanye Ma, Mark Goldstein, Michael S. Albergo, Nicholas M. Boffi, Eric Vanden-Eijnden, and Saining Xie. Sit: Exploring flow and diffusion-based generative models with scalable interpolant transformers. In ECCV (77), volume 15135 ofLecture Notes in Computer Science, pages 23–40....

    2024

  47. [55]

    Soda: Bottleneck diffusion models for representation learning

    Drew A Hudson, Daniel Zoran, Mateusz Malinowski, Andrew K Lampinen, Andrew Jaegle, James L McClelland, Loic Matthey, Felix Hill, and Alexander Lerchner. Soda: Bottleneck diffusion models for representation learning. InProceedings of the IEEE/CVF Conference on Computer Vision a...

    2024

  48. [56]

    Finite scalar quantization: Vq-vae made simple.arXiv preprint arXiv:2309.15505, 2023

    Fabian Mentzer, David Minnen, Eirikur Agustsson, and Michael Tschannen. Finite scalar quantization: Vq-vae made simple.arXiv preprint arXiv:2309.15505, 2023. 6

    2023 arXiv

  49. [57]

    Language model beats diffusion–tokenizer is key to visual generation.arXiv preprint arXiv:2310.05737, 2023

    Lijun Yu, José Lezama, Nitesh B Gundavarapu, Luca Versari, Kihyuk Sohn, David Minnen, Yong Cheng, Vighnesh Birodkar, Agrim Gupta, Xiuye Gu, et al. Language model beats diffusion–tokenizer is key to visual generation.arXiv preprint arXiv:2310.05737, 2023. 6

    arXiv

  50. [58]

    Courville

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron C. Courville. Film: Visual reasoning with a general conditioning layer. InAAAI, pages 3942–3951. AAAI Press, 2018. 6

    2018

  51. [59]

    simple diffusion: End-to-end diffusion for high resolution images

    Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. simple diffusion: End-to-end diffusion for high resolution images. InICML, volume 202 ofProceedings of Machine Learning Research, pages 13213– 13232. PMLR, 2023. 6

    2023

  52. [60]

    Simpler diffusion (sid2): 1.5 FID on imagenet512 with pixel-space diffusion.arXiv, abs/2410.19324, 2024

    Emiel Hoogeboom, Thomas Mensink, Jonathan Heek, Kay Lamerigts, Ruiqi Gao, and Tim Salimans. Simpler diffusion (sid2): 1.5 FID on imagenet512 with pixel-space diffusion.arXiv, abs/2410.19324, 2024. 6 18

    2024 arXiv

  53. [61]

    Efros, Eli Shechtman, and Oliver Wang

    Richard Zhang, Phillip Isola, Alexei A. Efros, Eli Shechtman, and Oliver Wang. The unreasonable effec- tiveness of deep features as a perceptual metric. InCVPR, pages 586–595. Computer Vision Foundation / IEEE Computer Society, 2018. 6

    2018

  54. [62]

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

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffusion transformers is easier than you think. arXiv, abs/2410.06940, 2024. 6

    arXiv

  55. [63]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, and Robin Rombach. Scaling rectified flow transformers for high-resolution image synthes...

    2024

  56. [64]

    Classifier-free diffusion guidance.arXiv, abs/2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv, abs/2207.12598, 2022. 7

    2022 arXiv

  57. [65]

    Root mean square layer normalization

    Biao Zhang and Rico Sennrich. Root mean square layer normalization. InNeurIPS, pages 12360–12371,

  58. [66]

    GLU variants improve transformer.arXiv, abs/2002.05202, 2020

    Noam Shazeer. GLU variants improve transformer.arXiv, abs/2002.05202, 2020. 7

    2002 arXiv

  59. [67]

    Jianlin Su, Murtadha H. M. Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024. 7, 10

    2024

  60. [68]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. InICLR, 2015. 7

    2015

  61. [69]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. 7

    2017 arXiv

  62. [70]

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

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. InNIPS, pages 6626–6637,

  63. [71]

    Improved techniques for training gans.Advances in neural information processing systems, 29, 2016

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans.Advances in neural information processing systems, 29, 2016. 7

    2016

  64. [72]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat gans on image synthesis. In NeurIPS, pages 8780–8794, 2021. 7

    2021

  65. [73]

    With an Explanation of the Method of Integration Employed in Constructing the Tables Which Give the Theoretical Forms of Such Drops

    Francis Bashforth and John Couch Adams.An Attempt to Test the Theories of Capillary Action by Comparing the Theoretical and Measured Forms of Drops of Fluid. With an Explanation of the Method of Integration Employed in Constructing the Tables Which Give the Theoretical Forms o...

  66. [74]

    Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, and William T. Freeman. Maskgit: Masked generative image transformer. InCVPR, pages 11305–11315. IEEE, 2022. 8

    2022

  67. [75]

    Scalable image tokenization with index backpropagation quantization.arXiv, 2024

    Fengyuan Shi, Zhuoyan Luo, Yixiao Ge, Yujiu Yang, Ying Shan, and Limin Wang. Scalable image tokenization with index backpropagation quantization.arXiv, 2024. 8

    2024

  68. [76]

    Lei Jimmy Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization.arXiv, abs/1607.06450,

    arXiv

  69. [77]

    Sigmoid-weighted linear units for neural network function approximation in reinforcement learning.Neural Networks, 107:3–11, 2018

    Stefan Elfwing, Eiji Uchibe, and Kenji Doya. Sigmoid-weighted linear units for neural network function approximation in reinforcement learning.Neural Networks, 107:3–11, 2018. 10

    2018

  70. [78]

    Query-key normaliza- tion for transformers

    Alex Henry, Prudhvi Raj Dachapally, Shubham Shantaram Pawar, and Yuxuan Chen. Query-key normaliza- tion for transformers. InEMNLP (Findings), volume EMNLP 2020 ofFindings of ACL, pages 4246–4253. Association for Computational Linguistics, 2020. 10 19

    2020

Pith tools

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