Pith. sign in

REVIEW 3 major objections 6 minor 51 references

Improving Progressive Generation with Decomposable Flow Matching

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read This paper establishes that progressive coarse-to-fine generation can be implemented inside a single Flow Matching model by giving each level of a multiscale representation its own flow timestep, and that this improves image and video…

desk verdict DFM is a simple, well-ablated progressive generation method whose FDD headline overstates the gain; the FID/IS evidence is thinner but consistent, and the paper deserves a serious referee. read the letter →

arxiv 2506.19839 v1 pith:7NYTRQJ7 submitted 2025-06-24 cs.CV cs.AI

Moayed Haji-Ali , Willi Menapace , Ivan Skorokhodov , Arpit Sahni , Sergey Tulyakov , Vicente Ordonez , Aliaksandr Siarohin This is my paper · ORCID
classification cs.CVcs.AI
keywords decomposableflowmatchingprogressivegenerationlaplacianpyramidmulti-scalespectralautoregressiondiffusiontransformerimageandvideo
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

Generating high-dimensional images and videos is expensive, and the paper's claim is that the task becomes both cheaper and higher-quality if the signal is generated coarse-to-fine inside an ordinary Flow Matching pipeline. Decomposable Flow Matching (DFM) decomposes the input into a multiscale representation, typically a Laplacian pyramid, assigns each level its own flow timestep $t_s$, and trains one shared model to predict per-scale velocities. At inference the model generates scales sequentially, starting with coarse structure and adding fine detail only once earlier scales reach a threshold, using a standard sampler. On ImageNet-1K at 512px the paper reports 35.2% FDD improvement over base Flow Matching and 26.4% over the best prior multi-stage baseline under equal training compute, with similar gains at 1024px, on Kinetics-700 video, and in finetuning a large text-to-image model. The wider point is that progressive generation can live in the scheduling of per-stage timesteps rather than in a new generative model family.

What carries the argument

The load-bearing object is the per-stage flow timestep vector $(t_1,\dots,t_S)$, which breaks the usual single global timestep into one per level of a multiscale decomposition and thereby turns a Flow Matching model into a coarse-to-fine progressive generator. The decomposition itself is a Laplacian pyramid, in which each level holds the difference between successive downsampled approximations, so levels contain progressively finer detail. Around these two objects the method adds per-scale patchification and timestep-embedding layers inside a DiT-style transformer so one network handles all scales, a masking term that suppresses loss from future fully-noisy scales, and a scale-equivariant autoencoder that maps RGB frequency bands to matching latent frequency bands. These pieces together replace stage-transition machinery with a simple schedule of per-scale timesteps.

What would settle it

Run the ImageNet-1K 512px comparison at fixed training compute with an autoencoder whose latent frequency bands are permuted or otherwise misaligned with RGB frequencies; if DFM's FDD advantage over vanilla Flow Matching collapses or reverses, the mechanism carrying the result is the latent spectral structure rather than the per-stage timestep schedule itself.

Watch

Extended reading notes

Core claim

The central claim is that Flow Matching and progressive generation can be fused without special diffusion processes, separate expert models, or ad-hoc samplers. The input is decomposed into $S$ scales $X^s$; each scale receives an independent flow timestep $t_s$, so the forward process is $X^s_{t_s} = t_s X^s_1 + (1-t_s) X^s_0$ per scale, and the model predicts per-scale velocities $v^s$ from the noisy versions of all scales. Training simulates the progressive schedule by sampling a current stage, drawing its timestep from a logit-normal distribution, drawing less-noisy timesteps for previous stages, setting future stages to full noise, and masking those future scales out of the loss. Inference advances the stages one by one with a standard ODE sampler, moving to the next scale once the current one reaches a threshold $\tau$. On ImageNet-1K, Kinetics-700, and a large-model finetuning task, the paper reports that this single-model scheme outperforms vanilla Flow Matching, cascaded diffusion, and Pyramidal Flow at matched training compute.

Load-bearing premise

The large quality gains depend on a compression model whose internal frequency bands line up with the visual frequency bands in RGB; without that spectral alignment the paper's own appendix shows DFM still beats Flow Matching, but the margin shrinks.

Editorial extensions

If this is right

  • On ImageNet-1K 512px, DFM reports a 35.2% FDD improvement over the base Flow Matching model and 26.4% over the best-performing prior baseline under equal training compute.
  • At 1024px and on Kinetics-700 video, DFM also beats Flow Matching, cascaded models, and Pyramidal Flow on FID, FDD, and related metrics at matched compute.
  • Finetuning a large text-to-image model with DFM reaches 28.7% lower FID and 24.3% lower FDD than standard full finetuning after the same number of steps, with higher CLIP similarity.
  • Because each scale is generated with its own timestep, intermediate stages can be decoded during inference, giving low-resolution previews without any extra model.
  • The decomposition is user-defined, so DWT, DCT, Fourier, or multiscale-autoencoder decompositions are drop-in alternatives to the Laplacian pyramid used in the main results.

Reading between the lines

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

  • If the per-stage timestep schedule is the active ingredient, the same factorization could be applied to non-diffusion generative models, e.g., next-scale autoregressive models, to test whether progressive structure itself or the flow formulation drives the gain.
  • The hand-set switch threshold and per-stage step counts suggest a natural extension: learn the sampling schedule along with the model, which might reach the same quality with fewer total steps.
  • For video, the same decomposition idea can be applied along time as well as space, which would test whether coarse-to-fine generation improves temporal consistency the way it appears to improve spatial structure.
  • A decisive comparison would use a spectrally disentangled tokenizer (e.g., wavelet-based) instead of a scale-equivariant fine-tuned autoencoder, isolating how much of the benefit comes from the latent structure versus the progressive schedule.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. Decomposable Flow Matching (DFM) extends Flow Matching to progressive multi-stage generation. The input latent is decomposed with a Laplacian pyramid into S scales; each scale receives its own flow timestep, and a shared DiT with per-scale patchification, per-scale time embeddings, and masked losses is trained to predict per-scale velocities. At inference, scales are denoised sequentially from coarse to fine according to a threshold schedule with per-stage step counts. Experiments on ImageNet-1K 512/1024 and Kinetics-700 compare DFM with Flow Matching, a cascaded model, and Pyramidal Flow under matched training compute; the paper reports consistent gains, e.g., FDD 200.6 vs 282.9 for Flow Matching on ImageNet 512 at cfg=1.0, and a 35.2% relative FDD improvement over the base architecture is quoted in the abstract. An additional experiment fine-tunes FLUX-dev on an internal dataset and reports faster convergence than standard fine-tuning. The appendix contains extensive ablations over training and sampling hyperparameters, decomposition choices, failed experiments, and limitations.

Significance. The framework is simple, decomposition-agnostic, single-model, and clearly described; these are genuine strengths. The ablation study is unusually thorough, the baseline comparisons are matched in training compute, and the convergence curves show the gain appears throughout training rather than only at the final checkpoint. The candor about failed experiments and limitations is helpful. However, the paper's quantitative claims would be more convincing with repeated seeds, a held-out validation protocol for hyperparameters, and a metric or human study that is sensitive to the high-frequency artifacts the authors acknowledge; the method's main reported gain is on FDD, a DINOv2-based structural metric. The lack of code and the private FLUX dataset also limit independent verification.

major comments (3)
  1. [§4.2, §4.3, Table 1, Figure 3] The main ImageNet-1K 512px configuration is tuned on the same benchmark used for the headline numbers: p0_t=0.9 and logit-normal location 1.5 are selected from Table 1(a)-(b), and threshold 0.7 with 30/10 per-stage steps is selected from Figure 3 and §4.3. With no validation split and no repeated-seed variance, part of the 35.2% FDD margin in Table 2 may be a selection effect rather than an intrinsic property of DFM. Please evaluate the ablated configurations on a held-out split, or report the sensitivity of the Table 2 comparisons to the selected hyperparameters.
  2. [§4.4, Table 2] All main results are single runs without error bars. Several margins are small enough to be within typical run-to-run variability, notably Kinetics-700 at cfg=1.0 (FDD 336.7 vs 353.8 for Pyramidal Flow, FVD 260.2 vs 265.6) and ImageNet 512 at cfg=1.5 (FID 4.28 vs 4.57 for Pyramidal Flow). The claim of uniform superiority over all baselines on all metrics therefore needs at least a few seeds for the main comparisons, or bootstrap confidence intervals on the evaluation metrics, to establish that the differences are not noise.
  3. [§5, Appendix G, Table 2] The abstract's headline '35.2% improvement in FDD' is a Frechet distance on DINOv2 features, which are biased toward global structure, while the paper itself reports 'a decreased presence of high-frequency details' and artifacts in vegetation, fur, and thin structures (§5, App. G, Fig. 7). The FDD gain may therefore partly reflect a shift toward the structural prior encoded by DINOv2 rather than an across-the-board improvement in visual quality. Because FID and IS also improve, I do not see this as invalidating the method, but the 'improves visual quality' claim should be qualified, or supported by a human evaluation or a texture-sensitive metric.
minor comments (6)
  1. [Eq. (4)] The mask term M_s in Eq. (4) is central to the training loss but is never defined in the main text; please specify which stages are masked and how M_s relates to the architectural input masking described in §3.4.
  2. [§4.4, Table 2] The sentence 'with the exception of Kinetics-700 FDD' is imprecise: at cfg=1.0 and 1.25 DFM has the best FDD on Kinetics-700; only at cfg=1.5 is DFM slightly behind Cascaded (236.5 vs 235.0).
  3. [Appendix A.1 / Table 1(c)] The main text says that removing standardization improves performance and that the main experiments exclude standardization, but Appendix A.1 says the model normalizes data across scales with scale-wise pre/post-conditioning; please clarify how the 'standardization' ablated in Table 1(c) differs from the preconditioning described in Appendix A.1.
  4. [§4.1, evaluation metrics] FDD is cited as 'Frechet Distance computed against DINOv2 features' but no details are given on the DINOv2 variant, layer, pooling, or number of samples used for the covariance estimate; please add these details.
  5. [Overall] The paper does not release code, and the FLUX experiment in §4.5 is on a private internal dataset; this limits independent verification of the empirical claims, especially given the absence of error bars.
  6. [Throughout] There are a few typographical errors, including 'add-hoc sampler' in the abstract and 'Addtionally' in Appendix C; please proofread.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: DFM is an empirical extension of Flow Matching; the only self-citations (autoencoder [38], training schedule [29]) are non-load-bearing, while the admitted high-frequency artifacts and same-benchmark hyperparameter selection raise evaluation-fairness concerns, not circularity.

full rationale

The derivational content of the paper (Sec. 3) is self-contained: Eq. (3) defines per-stage flow timesteps as independent linear interpolations, Eq. (4) is the corresponding multi-scale velocity loss, and Sec. 3.4 describes per-scale patchification and timestep embedders. No equation is defined in terms of the headline FDD/FID numbers, and no fitted constant is renamed as a prediction; the improvements in Table 2 are empirical comparisons run under matched training compute. The paper cites the authors' prior work for the scale-equivariant autoencoder ([38]) and for a cascaded training strategy ([29]), but both are used as engineering components or supporting intuition, equally for baselines and the proposed method; Appendix D additionally shows DFM still outperforms Flow Matching without scale-equivariant fine-tuning, so the self-citations are not load-bearing. The claimed limitations (Sec. 5 and App. G: a decreased presence of high-frequency details and artifacts in vegetation, fur, and thin structures) and the choice of DINOv2-based FDD as the headline metric are substantive threats to the strength of the 'improves visual quality' claim, but they do not make the derivation circular. Likewise, choosing p0_t=0.9, logit-normal location 1.5, threshold 0.7, and 30/10 sampling steps from ablations on ImageNet-1K 512px and then reporting Table 2 on the same benchmark is a model-selection caveat, not an equation-level reduction of the reported gain to its inputs by construction. Overall, the central framework has independent content, and the only circularity-adjacent issue is a minor, non-load-bearing self-citation.

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

The central claim relies on standard flow matching, Laplacian decomposition, and empirical training choices. The main non-standard assumption is that the autoencoder latents are spectrally structured, which is established by fine-tuning. The method also depends on several hyperparameters (stage probability, logit location, threshold, step allocation, decomposition resolutions) that are fitted via ablations on the same benchmarks used for the headline results.

free parameters (5)
  • First stage sampling probability p0_t = 0.9
    Controls how often the coarse scale is the focus during training; selected by ablations on ImageNet-1K 512px (Table 1a).
  • Preceding stage logit-normal location = 1.5
    Shifts earlier-stage timestep sampling toward low noise to mimic inference; chosen from Table 1b.
  • Sampling threshold tau = 0.7 for 512px, 0.95 for 1024px
    Determines when the next stage starts; tuned per resolution using FID/FDD (Section 4.3, Figure 3).
  • Per-stage sampling steps = 30 steps for stage 1, 10 steps for stage 2
    Allocation of 40 total Euler steps between scales; selected from sampling ablation (Section 4.3).
  • Stage resolution decomposition = 256->512 for 512px, 256->1024 for 1024px
    2-stage Laplacian decomposition; chosen from Table 1f ablation on ImageNet-1K 1024px.
assumptions (4)
  • standard math Laplacian pyramid levels sum to an approximation of the original signal (Eq. 2).
    Standard multiscale signal decomposition; reconstruction by summing levels.
  • domain assumption Scale-equivariant finetuning of the autoencoder yields latents whose frequency bands align with RGB frequency bands.
    Invoked in Section 2.3; used for all main experiments; without it the progressive decomposition may be less effective.
  • domain assumption A shared DiT with per-scale patchification and summed token embeddings can represent all stages without destructive interference.
    Validated empirically through ablations (Table 1c,d) but no formal guarantee.
  • domain assumption The per-scale flow matching loss with masking (Eq. 4) provides a valid training signal for progressive generation.
    No theoretical proof is given; the paper relies on empirical success.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Progressive Generation with Decomposable Flow Matching." pith.science (2026). https://pith.science/paper/7NYTRQJ7

@misc{pith2026250619839,
  author       = {Pith},
  title        = {Pith review of: Improving Progressive Generation with Decomposable Flow Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7NYTRQJ7}},
  note         = {Machine review of arXiv:2506.19839}
}
read the original abstract

Generating high-dimensional visual modalities is a computationally intensive task. A common solution is progressive generation, where the outputs are synthesized in a coarse-to-fine spectral autoregressive manner. While diffusion models benefit from the coarse-to-fine nature of denoising, explicit multi-stage architectures are rarely adopted. These architectures have increased the complexity of the overall approach, introducing the need for a custom diffusion formulation, decomposition-dependent stage transitions, add-hoc samplers, or a model cascade. Our contribution, Decomposable Flow Matching (DFM), is a simple and effective framework for the progressive generation of visual media. DFM applies Flow Matching independently at each level of a user-defined multi-scale representation (such as Laplacian pyramid). As shown by our experiments, our approach improves visual quality for both images and videos, featuring superior results compared to prior multistage frameworks. On Imagenet-1k 512px, DFM achieves 35.2% improvements in FDD scores over the base architecture and 26.4% over the best-performing baseline, under the same training compute. When applied to finetuning of large models, such as FLUX, DFM shows faster convergence speed to the training distribution. Crucially, all these advantages are achieved with a single model, architectural simplicity, and minimal modifications to existing training pipelines.

Figures

Figures reproduced from arXiv: 2506.19839 by the authors.

Figure 1
Figure 1. Comparison of different diffusion frameworks. Cascaded models employ a separate model [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of Decomposable Flow Matching (a) Sampling procedure. A sampling schedule [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FID10K results ablating sampling configuration, threshold, and per-stage steps. process. We report the results in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (22 more)
Figure 4
Figure 4. Figure 4: Comparison of DFM against baselines on ImageNet-1K 512px [ [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: (Top) Qualitative results produced by FLUX [ [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FID10K, FDD10K and IS ablating sampling configuration, threshold, and per-stage steps on DiT-XL/2 trained with DFM on ImageNet-1K [5]. D Autoencoders without Scale Equivariance DFM explicitly decouples low- and high-frequency content into two successive stages. The fir…
Figure 7
Figure 7. Figure 7: (left) Selected samples highlighting failure cases generated by our framework trained on [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Convergence curves for different datasets and metrics comparing our framework to baselines [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Convergence curves for different datasets and metrics comparing our framework to baselines [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Convergence curves for different datasets and metrics comparing our framework to [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Comparison of DFM against baselines on DiT-XL trained on ImageNet-1K 512px [ [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Comparison of DFM against baselines on DiT-XL trained on ImageNet-1K 1024px [ [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Comparison of DFM against baselines on DiT-XL trained on Kinetics-700 [ [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: Comparison of DFM against baselines on DiT-XL trained on Kinetics-700 [ [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: Comparison of DFM against baselines on DiT-XL trained on Kinetics-700 [ [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Comparison of finetuning FLUX-DEV with DFM against standard full finetuning trained finetuned for 24k steps. Samples are generated with cfg 4.5. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: Comparison of finetuning FLUX-DEV with DFM against standard full finetuning trained finetuned for 24k steps. Samples are generated with cfg 4.5. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: Ablation of per-stage sampling steps on DiT-XL trained on ImageNet-1K 512px [ [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]
Figure 19
Figure 19. Figure 19: Ablation of the effect of cfg values and sampling threshold [PITH_FULL_IMAGE:figures/full_fig_p029_19.png]
Figure 20
Figure 20. Figure 20: Qualitative results from selected classes on ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p030_20.png]
Figure 21
Figure 21. Figure 21: Qualitative results from selected classes on ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p031_21.png]
Figure 22
Figure 22. Figure 22: Qualitative results from selected classes on ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p032_22.png]
Figure 23
Figure 23. Figure 23: Fully uncurated samples from ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p033_23.png]
Figure 24
Figure 24. Figure 24: Fully uncurated samples from ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p034_24.png]
Figure 25
Figure 25. Figure 25: Fully uncurated samples from ImageNet-1K [ [PITH_FULL_IMAGE:figures/full_fig_p035_25.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 15 canonical work pages

  1. [1]

    Edify image: High-quality image generation with pixel space laplacian diffusion models.arXiv preprint arXiv:2411.07126, 2024

    Yuval Atzmon, Maciej Bala, Yogesh Balaji, Tiffany Cai, Yin Cui, Jiaojiao Fan, Yunhao Ge, Siddharth Gururani, Jacob Huffman, Ronald Isaac, et al. Edify image: High-quality image generation with pixel space laplacian diffusion models.arXiv preprint arXiv:2411.07126, 2024

    arXiv 2024

  2. [2]

    One transformer fits all distributions in multi-modal diffusion at scale

    Fan Bao, Shen Nie, Kaiwen Xue, Chongxuan Li, Shi Pu, Yaole Wang, Gang Yue, Yue Cao, Hang Su, and Jun Zhu. One transformer fits all distributions in multi-modal diffusion at scale. InInternational Conference on Machine Learning, pages 1692–1717. PMLR, 2023

    2023

  3. [3]

    A short note on the kinetics- 700 human action dataset.arXiv preprint arXiv:1907.06987, 2019

    Joao Carreira, Eric Noland, Chloe Hillier, and Andrew Zisserman. A short note on the kinetics- 700 human action dataset.arXiv preprint arXiv:1907.06987, 2019

    arXiv 1907

  4. [4]

    Generative pretraining from pixels

    Mark Chen, Alec Radford, Rewon Child, Jeff Wu, Heewoo Jun, David Luan, and Ilya Sutskever. Generative pretraining from pixels. InICML. JMLR.org, 2020

    work page 2020

  5. [5]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2009

    work page 2009

  6. [6]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12873–12883, 2021

    2021

  7. [7]

    Scaling rectified flow transform- ers 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, et al. Scaling rectified flow transform- ers for high-resolution image synthesis. InICML, 2024

    2024

  8. [8]

    Spectral image tokenizer.arXiv preprint arXiv:2412.09607, 2024

    Carlos Esteves, Mohammed Suhail, and Ameesh Makadia. Spectral image tokenizer.arXiv preprint arXiv:2412.09607, 2024

    arXiv 2024

Show all 51 references
  1. [9]

    f-dm: A multi-stage diffusion model via progressive signal transformation.ICLR (ICLR), 2023

    Jiatao Gu, Shuangfei Zhai, Yizhe Zhang, Miguel Angel Bautista, and Josh Susskind. f-dm: A multi-stage diffusion model via progressive signal transformation.ICLR (ICLR), 2023

    2023

  2. [10]

    Matryoshka diffusion models

    Jiatao Gu, Shuangfei Zhai, Yizhe Zhang, Joshua M Susskind, and Navdeep Jaitly. Matryoshka diffusion models. InThe Twelfth International Conference on Learning Representations, 2023

    2023

  3. [11]

    Av-link: Temporally-aligned diffusion features for cross-modal audio-video generation.arXiv preprint arXiv:2412.15191, 2024

    Moayed Haji-Ali, Willi Menapace, Aliaksandr Siarohin, Ivan Skorokhodov, Alper Canberk, Kwot Sin Lee, Vicente Ordonez, and Sergey Tulyakov. Av-link: Temporally-aligned diffusion features for cross-modal audio-video generation.arXiv preprint arXiv:2412.15191, 2024

    2024 arXiv

  4. [12]

    Classifier-free diffusion guidance.NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.NeurIPS 2021 Workshop on Deep Generative Models and Downstream Applications, 2022

    2021

  5. [13]

    Cascaded diffusion models for high fidelity image generation.Journal of Machine Learning Research, 23(47):1–33, 2022

    Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffusion models for high fidelity image generation.Journal of Machine Learning Research, 23(47):1–33, 2022

    2022

  6. [14]

    Video diffusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022

    Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video diffusion models.Advances in Neural Information Processing Systems, 35:8633–8646, 2022

    2022

  7. [15]

    Lora: Low-rank adaptation of large language models.ICLR, 1 (2):3, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 1 (2):3, 2022

    2022

  8. [16]

    Nfig: Autoregressive image generation with next-frequency prediction.arXiv preprint arXiv:2503.07076, 2025

    Zhihao Huang, Xi Qiu, Yukuo Ma, Yifu Zhou, Junjie Chen, Hongyuan Zhang, Chi Zhang, and Xuelong Li. Nfig: Autoregressive image generation with next-frequency prediction.arXiv preprint arXiv:2503.07076, 2025

    2025

  9. [17]

    Flexvar: Flexible visual autoregressive modeling without residual prediction.arXiv preprint arXiv:2502.20313, 2025

    Siyu Jiao, Gengwei Zhang, Yinlong Qian, Jiancheng Huang, Yao Zhao, Humphrey Shi, Lin Ma, Yunchao Wei, and Zequn Jie. Flexvar: Flexible visual autoregressive modeling without residual prediction.arXiv preprint arXiv:2502.20313, 2025. 11

    2025

  10. [18]

    Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954, 2024

    Yang Jin, Zhicheng Sun, Ningyuan Li, Kun Xu, Hao Jiang, Nan Zhuang, Quzhe Huang, Yang Song, Yadong Mu, and Zhouchen Lin. Pyramidal flow matching for efficient video generative modeling.arXiv preprint arXiv:2410.05954, 2024

    2024

  11. [19]

    Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35: 26565–26577, 2022

    Tero Karras, Miika Aittala, Timo Aila, and Samuli Laine. Elucidating the design space of diffusion-based generative models.Advances in neural information processing systems, 35: 26565–26577, 2022

    2022

  12. [20]

    Analyzing and improving the training dynamics of diffusion models

    Tero Karras, Miika Aittala, Jaakko Lehtinen, Janne Hellsten, Timo Aila, and Samuli Laine. Analyzing and improving the training dynamics of diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 24174–24184, 2024

    2024

  13. [21]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  14. [22]

    Eq-vae: Equivariance regularized latent space for improved generative image modeling.arXiv preprint arXiv:2502.09509, 2025

    Theodoros Kouzelis, Ioannis Kakogeorgiou, Spyros Gidaris, and Nikos Komodakis. Eq-vae: Equivariance regularized latent space for improved generative image modeling.arXiv preprint arXiv:2502.09509, 2025

    2025 arXiv

  15. [23]

    Flux, 2024

    Black Forest Labs. Flux, 2024. URLhttps://bfl.ai/

    2024

  16. [24]

    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, 2022

    2022

  17. [25]

    Hunyuan-dit: A powerful multi-resolution diffusion transformer with fine-grained chinese understanding.arXiv, 2024

    Zhimin Li, Jianwei Zhang, Qin Lin, Jiangfeng Xiong, Yanxin Long, Xinchi Deng, Yingfang Zhang, Xingchao Liu, Minbin Huang, Zedong Xiao, Dayou Chen, Jiajun He, Jiahao Li, Wenyue Li, Chen Zhang, Rongwei Quan, Jianxiang Lu, Jiabin Huang, Xiaoyan Yuan, Xiaoxiao Zheng, Yixuan Li, Ji...

    2024

  18. [26]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    2022 arXiv

  19. [27]

    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, 2023

    2023

  20. [28]

    Step-video-t2v technical report: The practice, challenges, and future of video foundation model.arXiv preprint arXiv:2502.10248, 2025

    Guoqing Ma, Haoyang Huang, Kun Yan, Liangyu Chen, Nan Duan, Shengming Yin, Changyi Wan, Ranchen Ming, Xiaoniu Song, Xing Chen, et al. Step-video-t2v technical report: The practice, challenges, and future of video foundation model.arXiv preprint arXiv:2502.10248, 2025

    2025 arXiv

  21. [29]

    Snap video: Scaled spatiotemporal transformers for text-to-video synthesis

    Willi Menapace, Aliaksandr Siarohin, Ivan Skorokhodov, Ekaterina Deyneka, Tsai-Shien Chen, Anil Kag, Yuwei Fang, Aleksei Stoliar, Elisa Ricci, Jian Ren, et al. Snap video: Scaled spatiotemporal transformers for text-to-video synthesis. InCVPR, 2024

    2024

  22. [30]

    Dctdiff: Intriguing properties of image generative modeling in the dct space.arXiv preprint arXiv:2412.15032, 2024

    Mang Ning, Mingxiao Li, Jianlin Su, Haozhe Jia, Lanmiao Liu, Martin Beneš, Wenshuo Chen, Albert Ali Salah, and Itir Onal Ertugrul. Dctdiff: Intriguing properties of image generative modeling in the dct space.arXiv preprint arXiv:2412.15032, 2024

    2024 arXiv

  23. [31]

    Conditional image generation with pixelcnn decoders

    Aäron van den Oord, Nal Kalchbrenner, Oriol Vinyals, Lasse Espeholt, Alex Graves, and Koray Kavukcuoglu. Conditional image generation with pixelcnn decoders. InNIPS, 2016

    2016

  24. [32]

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

    2024

  25. [33]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InICCV, 2023. 12

    2023

  26. [34]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InICML, 2021

    2021

  27. [35]

    Generating diverse high-fidelity images with vq-vae-2.Advances in neural information processing systems, 32, 2019

    Ali Razavi, Aaron Van den Oord, and Oriol Vinyals. Generating diverse high-fidelity images with vq-vae-2.Advances in neural information processing systems, 32, 2019

    2019

  28. [36]

    Generative modelling with inverse heat dissipation

    Severi Rissanen, Markus Heinonen, and Arno Solin. Generative modelling with inverse heat dissipation. InThe Eleventh International Conference on Learning Representations, 2023

    2023

  29. [37]

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

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding. InNeurIPS, 2022

    2022

  30. [38]

    Improving the diffusability of autoencoders

    Ivan Skorokhodov, Sharath Girish, Benran Hu, Willi Menapace, Yanyu Li, Rameen Abdal, Sergey Tulyakov, and Aliaksandr Siarohin. Improving the diffusability of autoencoders. In ICML, 2025

    2025

  31. [39]

    Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion models.Advances in Neural Information Processing Systems, 36:3732–3784, 2023

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

    2023

  32. [40]

    Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 2024

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

    2024

  33. [41]

    Relay diffusion: Unifying diffusion process across resolutions for image synthesis.arXiv preprint arXiv:2309.03350, 2023

    Jiayan Teng, Wendi Zheng, Ming Ding, Wenyi Hong, Jianqiao Wangni, Zhuoyi Yang, and Jie Tang. Relay diffusion: Unifying diffusion process across resolutions for image synthesis.arXiv preprint arXiv:2309.03350, 2023

    2023 arXiv

  34. [42]

    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

    2024

  35. [43]

    Towards accurate generative models of video: A new metric & challenges

    Thomas Unterthiner, Sjoerd Van Steenkiste, Karol Kurach, Raphael Marinier, Marcin Michalski, and Sylvain Gelly. Towards accurate generative models of video: A new metric & challenges. arXiv preprint arXiv:1812.01717, 2018

    2018 arXiv

  36. [44]

    One rank at a time: Cascading error dynamics in sequential learning.arXiv preprint arXiv:2505.22602, 2025

    Mahtab Alizadeh Vandchali, Anastasios Kyrillidis, et al. One rank at a time: Cascading error dynamics in sequential learning.arXiv preprint arXiv:2505.22602, 2025

    2025 arXiv

  37. [45]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. Attention is all you need. InNeurIPS, 2017

    2017

  38. [46]

    Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, et al. Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314, 2025

    2025 arXiv

  39. [47]

    Tokenformer: Rethinking transformer scaling with tokenized model parameters.arXiv preprint arXiv:2410.23168, 2024

    Haiyang Wang, Yue Fan, Muhammad Ferjad Naeem, Yongqin Xian, Jan Eric Lenssen, Liwei Wang, Federico Tombari, and Bernt Schiele. Tokenformer: Rethinking transformer scaling with tokenized model parameters.arXiv preprint arXiv:2410.23168, 2024

    arXiv

  40. [48]

    Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al. Cogvideox: Text-to-video diffusion models with an expert transformer.arXiv preprint arXiv:2408.06072, 2024

    2024 arXiv

  41. [49]

    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, 2022

    2022

  42. [50]

    Scaling autoregressive models for content-rich text-to-image generation.Transactions on Machine Learning Research, 2022

    Jiahui Yu, Yuanzhong Xu, Jing Yu Koh, Thang Luong, Gunjan Baid, Zirui Wang, Vijay Vasudevan, Alexander Ku, Yinfei Yang, Burcu Karagol Ayan, Ben Hutchinson, Wei Han, Zarana Parekh, Xin Li, Han Zhang, Jason Baldridge, and Yonghui Wu. Scaling autoregressive models for content-ric...

    2022

  43. [192]

    Please note that since FLUX-DEVis distilled and post-trained on highly aesthetic images, its distribution differs from that of our internal data

    We drop the text conditioning10%of the time to enable classifier-free guidance. Please note that since FLUX-DEVis distilled and post-trained on highly aesthetic images, its distribution differs from that of our internal data. Therefore, a direct comparison with the original FL...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.