Pith. sign in

REVIEW 4 major objections 6 minor 51 references

LSSGen: Leveraging Latent Space Scaling in Flow and Diffusion for Efficient Text to Image Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read LSSGen speeds up text-to-image diffusion and flow models by 1.5x by upsampling latents instead of pixels, keeping quality comparable and beating pixel-space scaling by up to 246% TOPIQ.

desk verdict A practical latent-space upsampling method with a genuine contribution, but the headline 246% TOPIQ claim needs correction. read the letter →

arxiv 2507.16154 v1 pith:3XYVFVRJ submitted 2025-07-22 cs.CV cs.AI

Jyun-Ze Tang , Chih-Fan Hsu , Jeng-Lin Li , Ming-Ching Chang , Wei-Chao Chen This is my paper
classification cs.CVcs.AI
keywords latentspacescalingflowmatchingdiffusionmodelsprogressiveupsamplingefficienttext-to-imagegenerationnoisescheduleVAE-agnosticupsampler
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes LSSGen, a way to speed up text-to-image diffusion and flow-matching models by starting generation at low resolution in the model's compressed latent space and progressively upsampling the latent itself with a small trained upsampler, instead of converting back to pixels between stages. It claims this preserves structure and avoids the blur and distortion that pixel-space upscaling introduces when images are re-encoded into latents. On FLUX.1-dev, the method cuts 1024x1024 generation time by about 1.5x while keeping alignment scores close to baseline and often improving perceptual quality. The same VAE-dependent upsampler transfers across flow and diffusion backbones that share a VAE, so the approach is train-once and reusable.

What carries the argument

The load-bearing mechanism is a lightweight VAE-dependent ResNet upsampler (about 500K parameters) paired with a resolution-aware noise schedule. The central identity is the SNR rescaling rule: doubling both latent dimensions quadruples the pixel count, which the paper treats as reducing the signal-to-noise ratio by a factor of 4, so after upsampling the noise coefficient becomes $\sigma'_t = \frac{3}{4}\sigma_t$ and each new stage starts at $\sigma_{init} \approx 0.75$. A timestep shift $t_m = \frac{\sqrt{m/n}\, t_n}{1 + (\sqrt{m/n} - 1)t_n}$ redistributes denoising steps toward cheaper low-resolution stages. The upsampler is trained for a few epochs on a generic image-caption dataset with a frozen VAE, which lets it transfer across generative backbones that share that VAE.

What would settle it

Measure the actual signal-to-noise change produced by the latent upsampler on a fixed VAE: take a clean latent, upsample it with the trained upsampler, and compare the variance of the upsampled output against the original. If the ratio is not close to 1/4, the $\sigma_{init} = 0.75$ rule is an empirical choice rather than a derived optimum. A second check is to reproduce the $\sigma_{init}$ sweep of Table 4 on a different VAE and see whether perceptual quality still peaks near 0.75.

Watch

Extended reading notes

Core claim

The paper's central claim is that the coarse-to-fine denoising trajectory of diffusion and flow models can be reproduced in latent space by a resolution-autoregressive loop: denoise at a low resolution, upsample the latent with a small learned ResNet upsampler rather than pixels, inject Gaussian noise at a carefully chosen level, and continue denoising at the next resolution. Because the latent upsampler depends only on the VAE, the same trained module plugs into any diffusion or flow backbone that shares that VAE, so FLUX.1-dev, FLUX.1-schnell, SD3.5-m, SDXL, SD1.5, Playground-v2.5, and LCM-SDXL all run without architectural changes. The paper argues that this avoids the encode-decode round trip that makes pixel-space scaling blurry, and that the right noise injection keeps the pretrained denoiser on-distribution. With this pipeline, 1024x1024 generation runs about 1.5x faster than the full-resolution baseline while retaining comparable alignment and improving perceptual metrics, and pixel-space baselines are beaten by up to 246% TOPIQ at similar speeds.

Load-bearing premise

The load-bearing premise is that doubling a latent's width and height really cuts its signal-to-noise ratio by a factor of four, making $\sigma_{init} \approx 0.75$ the right restart noise; the paper states this as a heuristic approximation and selects the value experimentally in an ablation.

Editorial extensions

If this is right

  • At 1024x1024, LSSGen provides a 1.5x speedup on FLUX.1-dev and SD3.5-m while keeping GenEval close to baseline and improving CLIP-IQA and TOPIQ.
  • At 2048x2048, where computational cost grows quadratically with resolution, the relative gains are largest: FLUX.1-schnell's GenEval rises from 0.489 to 0.657 with a 1.1x speedup and better perceptual scores.
  • Schedule shifting is what converts the resolution hierarchy into wall-clock savings: on SD3.5-m it raises the speedup from 1.1x to 1.5x while also improving CLIP-IQA and TOPIQ by about 3%.
  • Because the upsampler attaches to the VAE rather than the generative backbone, the method composes with distilled few-step models; LCM-SDXL and FLUX.1-schnell both improve quality without losing the speed benefit.

Reading between the lines

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

  • The paper leaves implicit that the $\sigma_{init} = 0.75$ rule should be re-calibrated per VAE; the empirical SNR of each VAE's latents can be measured directly, which would turn the heuristic into a parameter-free schedule.
  • A stress test the paper does not run: on VAEs with very few latent channels, latent-space upsampling may over-smooth or over-sharpen, and pixel-space scaling could become competitive; this boundary would define where latent scaling stops being the better choice.
  • Because alignment tracks the starting resolution more than the upsampling process, LSSGen inherits the base model's alignment ceiling at the starting resolution; lowering the start to 256x256 trades alignment for speed, so applications with strict prompt fidelity should keep starts at 512x512.
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

4 major / 6 minor

Summary. The paper proposes LSSGen, a framework for accelerating text-to-image generation in diffusion and flow-based models by performing early denoising stages at lower resolution directly in latent space, using a lightweight ResNet-based latent upsampler and a noise compensation/schedule-shifting strategy. The authors evaluate on FLUX.1-dev/schnell, SD3.5-m, SDXL, SD1.5, Playground-v2.5, and LCM-SDXL at 1024^2 and 2048^2, comparing against MegaFusion, DiffuseHigh, Self-Cascade, and baselines, reporting speedups and quality metrics (GenEval, CLIP-IQA, TOPIQ, NIQE). The central claims are that LSSGen achieves 1.5x speedup with comparable or better quality, and up to 246% TOPIQ improvement over pixel-space scaling methods.

Significance. If the results are reproducible, LSSGen is a practical contribution to efficient high-resolution text-to-image generation. Its strengths include a lightweight (~500K parameter) VAE-dependent upsampler that is reusable across models sharing a VAE, evaluation across multiple architectures and resolutions, and a clear algorithm description in the supplementary material. The upsampler is trained on an external dataset (COCO), providing independent grounding for that component. The main limitations are the unsupported headline improvement figure and the flawed theoretical derivation of the initial noise coefficient, both of which are correctable in revision.

major comments (4)
  1. [Abstract and Conclusion] The abstract and conclusion state that LSSGen achieves up to 246% TOPIQ score improvement when generating 1024^2 images at similar speeds. Recomputing relative improvements from Table 1 gives a maximum of 75.4% against MegaFusion on FLUX.1-dev at 1024^2 (0.705 vs 0.402), or 15.8% against the FLUX.1-schnell 2048 baseline (0.586 vs 0.506), or 145.6% against MegaFusion++ on SD1.5 (0.609 vs 0.248) but at a 2.3x speed difference. None of these equals 246%. Please either provide the exact comparison basis for the 246% figure or correct the claim to match the reported data.
  2. [Sec. 4.1, Eqs. (7)-(8)] The derivation of the 'theoretical optimal' sigma_init ~ 0.75 is inconsistent with the SNR formulation. From Eq. (7), sigma = 1/(1+sqrt(SNR)). If upsampling reduces SNR to SNR/4, then sigma' = 1/(1+sqrt(SNR/4)) = 2*sigma/(1+sigma), not 3/4*sigma. Inserting sigma = 1 (pure noise) gives sigma' = 1, not 0.75. The text also invokes 'a linear relationship between SNR and sigma' that contradicts Eq. (8). The empirical ablation in Table 4 can support sigma_init = 0.75, but the paper should either present a correct derivation or explicitly label this value as an empirical choice.
  3. [Sec. 5.2 and Table 1] The introduction and conclusion claim a 1.5x speedup for 1024^2 generation, but Table 1 shows this only for FLUX.1-dev and SD3.5-m; the Playground-v2.5 and FLUX.1-schnell rows show 1.1x, and SD1.5 shows 1.0x. Please qualify the speedup claim according to the actual rows and avoid implying it holds across all evaluated models.
  4. [Sec. 4.2 and Sec. 5.1] The paper states the upsampler is VAE-dependent and reusable across models sharing the same VAE, but it does not specify which VAE(s) it was trained on or whether separate upsamplers were trained for the different models evaluated (FLUX, SD3.5-m, SDXL, SD1.5). Without this information, the 'train-once, use-across-models' claim cannot be verified. Please report the VAE used for training and any per-VAE variants, along with the number of upsamplers trained.
minor comments (6)
  1. [Throughout] The text frequently uses 'V AE' with a space; this should be 'VAE' (e.g., in the Introduction, Sec. 4.2, and Figure 2 caption).
  2. [Table 5] Table 5 reports results for two different backbones (SD3.5-m for the shift comparison and FLUX.1-dev for the step-shortening comparison) in the same table without a clear grouping; please split the table or add a backbone column.
  3. [Sec. 5.2] The statement 'LSSGen improves image quality by 3-8% over the baseline' should specify which metric and which baseline row; for the FLUX.1-dev row, TOPIQ improves from 0.674 to 0.705 (4.6%) and CLIP-IQA from 0.887 to 0.914 (3.0%).
  4. [Supplementary B.3] The supplementary material mentions a potential over-sharpening artifact for human faces, but this limitation is not mentioned in the main text; please add it to the limitations discussion.
  5. [Eq. (10)] Equation (10) is unclear as typeset; the shift factor appears to be sqrt(m/n) but is not defined explicitly. Please clarify the notation and define the variables.
  6. [Table 3] In Table 3, the 256x256 baseline row has an outlier NIQE value of 8.729; please add a note or explanation for this value.

Circularity Check

1 steps flagged · score 4.0 of 10

Sigma-init 'theoretical optimum' is post-hoc selected by the same ablation metrics, though the core latent-upsampling result has independent grounding.

  1. fitted input called prediction [Sec. 4.1 (Eqs. 7-9) and Sec. 5.3 (Table 4)]
    "For example, scaling the resolution by a factor of s = 2 ... we approximate as reducing SNR(t) to 1/4 SNR(t). ... we scale the noise coefficient after upsampling as σ′t = 3/4 σt ... At this scale, the signal-to-noise ratio is approximately 1/4 SNR, corresponding to a noise coefficient of σinit ≈ 0.75 ... As σinit approaches the theoretical optimal value of 0.75, perceptual quality improves steadily ... This supports our theoretical insight that σinit ≈ 0.75 is optimal for half-resolution initialization."

    The 'theoretical' value is not entailed by Eqs. 7-8: under SNR' = SNR/4, Eq. 8 gives σ' = 1/(1+√(SNR/4)) = 2σ/(1+σ), which equals 3σ/4 only at the impossible σ = 5/3, and at σ=1 it gives σ'=1, not 0.75. The paper even invokes 'a linear relationship between SNR and σ', but Eq. 8 is nonlinear. The number 0.75 is instead the value that the Table 4 sweep on SD3.5-m selects using GenEval, CLIP-IQA, TOPIQ, and NIQE, the same metrics and prompt set used for the paper's headline quality claims. Thus the 'theoretical optimal' label is a post-hoc interpretation of an ablated optimum, and the reported final quality numbers are generated with a hyperparameter chosen on that same benchmark rather than independently predicted.

full rationale

The central mechanism is not circular overall: the latent upsampler is trained on COCO with a frozen VAE, so its weights are external to the GenEval/TOPIQ benchmark; the timestep-shift formula (Eq. 10) is imported from Stable Diffusion 3 rather than derived from the paper's own claims; and the speed advantage follows from the O((H×W)^2) attention complexity of starting at low resolution. No load-bearing self-citation chain appears, and the external baselines (MegaFusion, Self-Cascade, DiffuseHigh) are reproduced from released code. The one genuinely circular-adjacent step is the 'theoretical optimal' σinit=0.75 in Sec. 4.1: the derivation is inconsistent with Eqs. 7-8, and the value is confirmed by the Table 4 ablation on the same metrics used for the final quality comparisons, so that particular claim is fitted rather than predicted. I also flag that the abstract/conclusion 'up to 246% TOPIQ score improvement' is not supported by Table 1: re-computing relative gains gives at most ~145.6% (LSS-SD1.5 vs SD1.5-MegaFusion++) or ~75.4% (LSS-FLUX.1-dev vs MegaFusion at 1024); this is a numerical-support/correctness defect, not a circular-reduction defect, so under the hard rules it does not raise the circularity score.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

The framework depends on a small number of tuned constants: the starting noise level sigma_init and the SNR reduction factor. The upsampler is the main learned component and provides some external grounding. The core heuristic assumptions are the blurry-to-sharp equivalence and the 1/4 SNR reduction, both stated without rigorous derivation.

free parameters (2)
  • sigma_init (initial noise coefficient) = 0.75 (optimal); range 0.3-1.0 tested
    Chosen by ablation (Table 4) and described as theoretical optimum despite the SNR scaling argument being a heuristic; it is a fitted hyperparameter that controls the speed-quality trade-off and is central to the reported improvements.
  • SNR reduction factor (1/4 and noise coefficient 3/4) = 1/4 SNR reduction, sigma'_t = 3/4 sigma_t
    Introduced as an approximation in Sec. 4.1 to justify sigma_init; not derived from first principles, effectively calibrated to the ablation results.
assumptions (4)
  • domain assumption Generative path from blurry to sharp mirrors a resolution-increasing process, so early denoising steps can be run at low resolution.
    Sec. 4.1 states this as the basis for the framework; it is a plausible heuristic but not proven for arbitrary prompts/models.
  • ad hoc to paper Upsampling by 2x in latent space reduces SNR by a factor of 4.
    Sec. 4.1: 'we approximate as reducing SN R(t) to 1/4 SN R(t)' with no derivation beyond a pixel-count analogy; this drives sigma_init = 0.75.
  • domain assumption The VAE encoder is shared across models and remains frozen; a single upsampler trained on COCO generalizes across models sharing the VAE.
    Sec. 4.2: this is the key to 'train-once use-across-models'; it depends on the empirical claim that VAE latents are model-agnostic.
  • standard math Timestep shifting formula (Eq. 10) improves efficiency by allocating more steps to low-resolution stages.
    Adopted from Stable Diffusion 3 (Esser et al.), cited; we treat it as prior work, though the paper does not re-derive it.
invented entities (1)
  • Latent upsampler U (ResNet, ~500K params) independent evidence
    purpose: Upsamples latent features by 2x in latent space to avoid pixel-space re-encoding artifacts.
    The upsampler is trained on COCO with a frozen VAE and evaluated in the LSSGen pipeline; its outputs are compared against alternatives in Table 2 and Fig. 5, providing empirical evidence. It is not a speculative entity, but a trained component.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LSSGen: Leveraging Latent Space Scaling in Flow and Diffusion for Efficient Text to Image Generation." pith.science (2026). https://pith.science/paper/3XYVFVRJ

@misc{pith2026250716154,
  author       = {Pith},
  title        = {Pith review of: LSSGen: Leveraging Latent Space Scaling in Flow and Diffusion for Efficient Text to Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3XYVFVRJ}},
  note         = {Machine review of arXiv:2507.16154}
}
abstract

Flow matching and diffusion models have shown impressive results in text-to-image generation, producing photorealistic images through an iterative denoising process. A common strategy to speed up synthesis is to perform early denoising at lower resolutions. However, traditional methods that downscale and upscale in pixel space often introduce artifacts and distortions. These issues arise when the upscaled images are re-encoded into the latent space, leading to degraded final image quality. To address this, we propose {\bf Latent Space Scaling Generation (LSSGen)}, a framework that performs resolution scaling directly in the latent space using a lightweight latent upsampler. Without altering the Transformer or U-Net architecture, LSSGen improves both efficiency and visual quality while supporting flexible multi-resolution generation. Our comprehensive evaluation covering text-image alignment and perceptual quality shows that LSSGen significantly outperforms conventional scaling approaches. When generating $1024^2$ images at similar speeds, it achieves up to 246\% TOPIQ score improvement.

Figures

Figures reproduced from arXiv: 2507.16154 by the authors.

Figure 1
Figure 1. Comparison of image synthesis results at [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our Latent Space Scaling Generation (LSSGen) framework. The inference process is divided into multiple stages, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Upsampler architecture. the pixel count by 4×, which we approximate as reducing SNR(t) to 1 4 SNR(t). This aligns with the view of bilinear interpolation as a smoothing process that effectively adds noise. To reflect this in the generative path, we scale the noise coefficient after upsampling as σ ′ t = 3 4 σt, providing a more suitable initialization for denoising and better aligning with the underlying probabilist… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visual comparison for different generation frameworks on SDXL with random GenEval prompts at resolution [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: A visual comparison of resolution scaling methods applied within the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

51 extracted references · 36 canonical work pages

  1. [1]

    FLUX.1: Text-to-image Generation Model

    Black Forest Labs. FLUX.1: Text-to-image Generation Model. https://github.com/black- forest- labs/flux, 2024. Released: August 2, 2024. 1, 3, 5, 6, 8, 9

    work page 2024

  2. [2]

    TOPIQ: A top-down approach from semantics to distortions for image quality assessment

    Chaofeng Chen, Jiadi Mo, Jingwen Hou, Haoning Wu, Liang Liao, Wenxiu Sun, Qiong Yan, and Weisi Lin. TOPIQ: A top-down approach from semantics to distortions for image quality assessment. IEEE Transactions on Image Processing,

    work page

  3. [3]

    Neural ordinary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. Advances in Neural Information Processing Systems , 31,

    work page

  4. [4]

    Emu: Enhancing Image Generation Models Using Photogenic Needles in a Haystack

    Xiaoliang Dai, Ji Hou, Chih-Yao Ma, Sam S Tsai, Jialiang Wang, Rui Wang, Peizhao Zhang, Simon Vandenhende, Xi- aofang Wang, Abhimanyu Dubey, et al. Emu: Enhancing Image Generation Models Using Photogenic Needles in a Haystack. CoRR, 2023. 7

    work page 2023

  5. [5]

    Flow matching in latent space

    Quan Dao, Hao Phung, Binh Nguyen, and Anh Tran. Flow matching in latent space. arXiv preprint arXiv:2307.08698,

    arXiv

  6. [6]

    Adjoint matching: Fine-tuning flow and diffusion generative models with memoryless stochastic op- timal control

    Carles Domingo-Enrich, Michal Drozdzal, Brian Karrer, and Ricky TQ Chen. Adjoint matching: Fine-tuning flow and diffusion generative models with memoryless stochastic op- timal control. arXiv preprint arXiv:2409.08861, 2024. 1

    arXiv 2024

  7. [7]

    Demofusion: Democratising high- resolution image generation with no $$$

    Ruoyi Du, Dongliang Chang, Timothy Hospedales, Yi-Zhe Song, and Zhanyu Ma. Demofusion: Democratising high- resolution image generation with no $$$. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6159–6168, 2024. 2

    work page 2024

  8. [8]

    I-max: Maximize the resolu- tion potential of pre-trained rectified flow transformers with projected flow

    Ruoyi Du, Dongyang Liu, Le Zhuo, Qin Qi, Hongsheng Li, Zhanyu Ma, and Peng Gao. I-max: Maximize the resolu- tion potential of pre-trained rectified flow transformers with projected flow. arXiv preprint arXiv:2410.07536, 2024. 3

    arXiv 2024

Show all 51 references
  1. [9]

    Taming transformers for high-resolution image synthesis

    Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. In Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12873–12883, 2021. 3

    2021

  2. [10]

    Scaling recti- fied flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M ¨uller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling recti- fied flow transformers for high-resolution image synthesis. In Proceedings of the International Conference ...

    2024

  3. [11]

    A Fourier Space Perspective on Diffu- sion Models

    Fabian Falck, Teodora Pandeva, Kiarash Zahirnia, Rachel Lawrence, Richard Turner, Edward Meeds, Javier Zazo, and Sushrut Karmalkar. A Fourier Space Perspective on Diffu- sion Models. arXiv preprint arXiv:2505.11278, 2025. 2, 4

    2025 arXiv

  4. [12]

    Geneval: An object-focused framework for evaluating text- to-image alignment

    Dhruba Ghosh, Hannaneh Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text- to-image alignment. Advances in Neural Information Pro- cessing Systems, 36:52132–52152, 2023. 2, 5

    2023

  5. [13]

    Smooth diffusion: Crafting smooth latent spaces in dif- fusion models

    Jiayi Guo, Xingqian Xu, Yifan Pu, Zanlin Ni, Chaofei Wang, Manushree Vasu, Shiji Song, Gao Huang, and Humphrey Shi. Smooth diffusion: Crafting smooth latent spaces in dif- fusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pag...

    2024

  6. [14]

    Make a cheap scaling: A self- cascade diffusion model for higher-resolution adaptation

    Lanqing Guo, Yingqing He, Haoxin Chen, Menghan Xia, Xi- aodong Cun, Yufei Wang, Siyu Huang, Yong Zhang, Xintao Wang, Qifeng Chen, et al. Make a cheap scaling: A self- cascade diffusion model for higher-resolution adaptation. In Proceedings of the European Conference on Compute...

    2024

  7. [15]

    Improved noise schedule for diffusion training

    Tiankai Hang, Shuyang Gu, Xin Geng, and Baining Guo. Improved noise schedule for diffusion training. arXiv preprint arXiv:2407.03297, 2024. 4

    2024 arXiv

  8. [16]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 770–778, 2016. 5

    2016

  9. [17]

    Denoising diffu- sion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffu- sion probabilistic models. Advances in Neural Information Processing Systems, 33:6840–6851, 2020. 3

    2020

  10. [18]

    Cascaded diffu- sion models for high fidelity image generation

    Jonathan Ho, Chitwan Saharia, William Chan, David J Fleet, Mohammad Norouzi, and Tim Salimans. Cascaded diffu- sion models for high fidelity image generation. Journal of Machine Learning Research, 23(47):1–33, 2022. 3

    2022

  11. [19]

    Diffusehigh: Training-free progressive high- resolution image synthesis through structure guidance

    Younghyun Kim, Geunmin Hwang, Junyu Zhang, and Eun- byung Park. Diffusehigh: Training-free progressive high- resolution image synthesis through structure guidance. In Proceedings of the AAAI Conference on Artificial Intelli- gence, pages 4338–4346, 2025. 2, 3, 5, 6

    2025

  12. [20]

    Variational diffusion models

    Diederik Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Variational diffusion models. Advances in Neural Infor- mation Processing Systems, 34:21696–21707, 2021. 4

    2021

  13. [21]

    Auto-encoding varia- tional Bayes

    Diederik P Kingma and Max Welling. Auto-encoding varia- tional Bayes. International Conference on Learning Repre- sentations, ICLR, 2014. 3

    2014

  14. [22]

    Playground v2.5: Three insights towards enhancing aesthetic quality in text-to-image genera- tion

    Daiqing Li, Aleks Kamko, Ehsan Akhgari, Ali Sabet, Lin- miao Xu, and Suhail Doshi. Playground v2.5: Three insights towards enhancing aesthetic quality in text-to-image genera- tion. arXiv preprint arXiv:2402.17245, 2024. 5, 6

    2024 arXiv

  15. [23]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll´ar, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In Proceedings of the European Conference on Computer Vi- sion, pages 740–755, 2014. 5

    2014

  16. [24]

    Flow Matching for Generative Modeling

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximil- ian Nickel, and Matthew Le. Flow Matching for Generative Modeling. In Proceedings of the International Conference on Learning Representations, 2023. 3

    2023

  17. [25]

    CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Trans- formers Up

    Songhua Liu, Zhenxiong Tan, and Xinchao Wang. CLEAR: Conv-Like Linearization Revs Pre-Trained Diffusion Trans- formers Up. arXiv preprint arXiv:2412.16112, 2024. 2

    2024 arXiv

  18. [26]

    Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow

    Xingchao Liu, Chengyue Gong, et al. Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow. In Proceedings of the International Conference on Learning Representations, 2023. 3

    2023

  19. [27]

    Latent consistency models: Synthesizing high- resolution images with few-step inference

    Simian Luo, Yiqin Tan, Longbo Huang, Jian Li, and Hang Zhao. Latent consistency models: Synthesizing high- resolution images with few-step inference. arXiv preprint arXiv:2310.04378, 2023. 2, 5, 6 9

    2023 arXiv

  20. [28]

    completely blind

    Anish Mittal, Rajiv Soundararajan, and Alan C Bovik. Mak- ing a “completely blind” image quality analyzer. IEEE Sig- nal Processing Letters, 20(3):209–212, 2012. 2, 6

    2012

  21. [29]

    Understanding the latent space of diffusion models through the lens of Riemannian geome- try

    Yong-Hyun Park, Mingi Kwon, Jaewoong Choi, Junghyo Jo, and Youngjung Uh. Understanding the latent space of diffusion models through the lens of Riemannian geome- try. Advances in Neural Information Processing Systems, 36: 24129–24142, 2023. 3

    2023

  22. [30]

    Explor- ing Diffusion and Flow Matching Under Generator Match- ing

    Zeeshan Patel, James DeLoye, and Lance Mathias. Explor- ing Diffusion and Flow Matching Under Generator Match- ing. arXiv preprint arXiv:2412.11024, 2024. 3

    2024 arXiv

  23. [31]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF Inter- national Conference on Computer Vision, pages 4195–4205,

  24. [32]

    SDXL: Improving Latent Diffusion Mod- els for High-Resolution Image Synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas M ¨uller, Joe Penna, and Robin Rombach. SDXL: Improving Latent Diffusion Mod- els for High-Resolution Image Synthesis. In Proceedings of the International Conference on Learning Representations ,

  25. [33]

    Boosting diffusion models with moving average sampling in frequency domain

    Yurui Qian, Qi Cai, Yingwei Pan, Yehao Li, Ting Yao, Qibin Sun, and Tao Mei. Boosting diffusion models with moving average sampling in frequency domain. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8911–8920, 2024. 2, 4

    2024

  26. [34]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In Proceedings of the International Conference ...

    2021

  27. [35]

    Stochastic backpropagation and approximate inference in deep generative models

    Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wier- stra. Stochastic backpropagation and approximate inference in deep generative models. In Proceedings of the Interna- tional Conference on Machine Learning , pages 1278–1286,

  28. [36]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj ¨orn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 10684–10695, 2022. 2, 5, 6

    2022

  29. [37]

    Progressive Distillation for Fast Sampling of Diffusion Models

    Tim Salimans and Jonathan Ho. Progressive Distillation for Fast Sampling of Diffusion Models. InProceedings of the In- ternational Conference on Learning Representations , 2022. 1, 2

    2022

  30. [38]

    FMBoost: Boosting Latent Diffusion with Flow Matching

    Johannes Schusterbauer, Ming Gui, Pingchuan Ma, Nick Stracke, Stefan Andreas Baumann, Vincent Tao Hu, and Bj¨orn Ommer. FMBoost: Boosting Latent Diffusion with Flow Matching. In Proceedings of the European Conference on Computer Vision, pages 338–355, 2024. 1, 3

    2024

  31. [39]

    Denois- ing Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denois- ing Diffusion Implicit Models. In Proceedings of the Inter- national Conference on Learning Representations, 2021. 3

    2021

  32. [40]

    Unraveling the Connections between Flow Matching and Diffusion Probabilistic Mod- els in Training-free Conditional Generation

    Kaiyu Song and Hanjiang Lai. Unraveling the Connections between Flow Matching and Diffusion Probabilistic Mod- els in Training-free Conditional Generation. arXiv preprint arXiv:2411.07625, 2024. 4

    2024 arXiv

  33. [41]

    Attention is all you need

    A Vaswani. Attention is all you need. Advances in Neural Information Processing Systems, 2017. 2, 1

    2017

  34. [42]

    Diffusers: State-of-the-art diffu- sion models

    Patrick von Platen, Suraj Patil, Anton Lozhkov, Pedro Cuenca, Nathan Lambert, Kashif Rasul, Mishig Davaadorj, Dhruv Nair, Sayak Paul, William Berman, Yiyi Xu, Steven Liu, and Thomas Wolf. Diffusers: State-of-the-art diffu- sion models. https://github.com/huggingface/ diffusers...

    2022

  35. [43]

    Ex- ploring clip for assessing the look and feel of images

    Jianyi Wang, Kelvin CK Chan, and Chen Change Loy. Ex- ploring clip for assessing the look and feel of images. InPro- ceedings of the AAAI Conference on Artificial Intelligence , pages 2555–2563, 2023. 2, 6

    2023

  36. [44]

    MegaFusion: Extend Dif- fusion Models towards Higher-resolution Image Generation without Further Tuning

    Haoning Wu, Shaocheng Shen, Qiang Hu, Xiaoyun Zhang, Ya Zhang, and Yanfeng Wang. MegaFusion: Extend Dif- fusion Models towards Higher-resolution Image Generation without Further Tuning. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV),...

    2025

  37. [45]

    FreeDiff: Progressive Frequency Trun- cation for Image Editing with Diffusion Models

    Wei Wu, Qingnan Fan, Shuai Qin, Hong Gu, Ruoyu Zhao, and Antoni B Chan. FreeDiff: Progressive Frequency Trun- cation for Image Editing with Diffusion Models. InProceed- ings of the European Conference on Computer Vision, pages 194–209, 2024. 2, 4

    2024

  38. [46]

    Sana: Efficient high-resolution image syn- thesis with linear diffusion transformers

    Enze Xie, Junsong Chen, Junyu Chen, Han Cai, Haotian Tang, Yujun Lin, Zhekai Zhang, Muyang Li, Ligeng Zhu, Yao Lu, et al. Sana: Efficient high-resolution image syn- thesis with linear diffusion transformers. arXiv preprint arXiv:2410.10629, 2024. 2

    arXiv

  39. [47]

    Deconvolutional networks

    Matthew D Zeiler, Dilip Krishnan, Graham W Taylor, and Rob Fergus. Deconvolutional networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2528–2535, 2010. 5

    2010

  40. [48]

    Dynamic diffusion transformer

    Wangbo Zhao, Yizeng Han, Jiasheng Tang, Kai Wang, Yib- ing Song, Gao Huang, Fan Wang, and Yang You. Dynamic diffusion transformer. arXiv preprint arXiv:2410.03456 ,

    arXiv

  41. [49]

    Designing a better asymmetric vqgan for stablediffusion

    Zixin Zhu, Xuelu Feng, Dongdong Chen, Jianmin Bao, Le Wang, Yinpeng Chen, Lu Yuan, and Gang Hua. Designing a better asymmetric vqgan for stablediffusion. arXiv preprint arXiv:2306.04632, 2023. 3 10 LSSGen: Leveraging Latent Space Scaling in Flow and Diffusion for Efficient Tex...

    2023 arXiv

  42. [50]

    Reducing patch count through our dynamic resolution strategy provides quadratic computational savings in self-attention operations

  43. [51]

    The quartic relationship between resolution and com- putational requirements (in both FLOPs and memory) demonstrates the theoretical efficiency of our progres- sive resolution approach compared to fixed-resolution methods. B. Case Study B.1. Effect of Initial Noise Intensity P...

Pith tools

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