Pith. sign in

REVIEW 3 major objections 5 minor 3 cited by

Exploring Representation-Aligned Latent Space for Better Generation

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that training a VAE whose latents are aligned with DINOv2 semantic features gives existing DiT and SiT diffusion models about a 15% FID improvement with no architecture changes, plus training-free segmentation and depth…

desk verdict Solid empirical recipe, but the paper does not isolate the effect it claims: no control VAE without the alignment loss, so the 15% FID gain could be KL tuning plus normalization. read the letter →

arxiv 2502.00359 v1 pith:P6SSMKL6 submitted 2025-02-01 cs.LG

Wanghan Xu , Xiaoyu Yue , Zidong Wang , Yao Teng , Wenlong Zhang , Xihui Liu , Luping Zhou , Wanli Ouyang
show 1 more author
Lei Bai
This is my paper · ORCID
classification cs.LG
keywords latentdiffusionmodelsvariationalautoencodersDINOv2semanticalignmentimagegenerationFIDspacetransformers
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

Latent diffusion models generate images in a compressed space produced by a VAE, and this paper argues that the semantic content of that space, not just its reconstruction fidelity, sets the ceiling on generation quality. ReaLS trains the VAE so that its latents, after a small MLP projection, reproduce the patch-level and global features of DINOv2, injecting semantic structure directly into the latent space. With DiT and SiT trained unchanged in this space, FID improves by about 15% on ImageNet 256x256, and the latents become semantically consistent enough to feed training-free segmentation and depth-estimation heads. The paper also shows that combining this latent-space alignment with feature-space alignment of the diffusion model improves FID further, reaching 18.96 at 1000k training steps.

What carries the argument

The key machinery is the ReaLS training objective plus a latent-rescaling procedure. The loss is $L = L_{\text{pixel}} + \lambda_k L_{\text{KL}} + \lambda_a L_{\text{align}}$, with $L_{\text{pixel}}$ combining MSE, GAN, and LPIPS losses and $L_{\text{align}}$ mixing a cosine-similarity term and a smooth-MSE term between MLP-projected latents and DINOv2 features, including both patch tokens and the global class token. During VAE training, a two-layer MLP alignment network maps each latent token and a pooled global token into DINOv2's feature dimension, forcing the latent space to mirror DINOv2's semantics; afterwards, latents are max-min normalized into the SD-VAE numerical range before diffusion training. The balance between KL and alignment, set with $\lambda_k = 2\times 10^{-5}$ and $\lambda_a = 1.0$, is what keeps the space both near standard normal and semantically structured.

What would settle it

Train a control VAE with the exact ReaLS recipe, including the same SD-VAE initialization, ImageNet data, pixel losses, KL weight $\lambda_k = 2\times 10^{-5}$, and max-min normalization, but with the alignment weight $\lambda_a = 0$; then train SiT-B/2 for 400k steps on both latent spaces. If the control's FID is close to ReaLS's 27.53 rather than near the SD-VAE baseline of 33.0, semantic alignment is not the cause of the improvement.

Watch

Extended reading notes

Core claim

The central claim is that a VAE whose latent space is aligned with DINOv2 semantics during training provides a better substrate for latent diffusion than a reconstruction-optimized VAE such as SD-VAE, without any change to the diffusion model. The authors call this latent space ReaLS and validate it by training DiT-B/2 and SiT-B/2, L/2, and XL/2; SiT-XL/2 plus ReaLS reaches an FID of 1.82 with classifier-free guidance after 400 epochs, while the no-guidance 400k-step FID improves from 43.5 to 35.27 for DiT-B/2 and from 33.0 to 27.53 for SiT-B/2. Reconstruction quality drops slightly relative to SD-VAE, which the paper reads as evidence that semantic richness matters more than reconstruction fidelity for generation. ReaLS latents also exhibit stronger category clustering in t-SNE, higher patch attention on same-object tokens, and higher semantic-consistency scores under image augmentations, and the alignment network lets downstream heads read semantics directly from generated latents.

Load-bearing premise

The improvement is credited to adding semantic alignment to the latent space, but the training recipe also relaxed the standard-normal constraint and rescaled the latents; the paper does not test a version with the same recipe but with the alignment turned off.

Editorial extensions

If this is right

  • Any latent diffusion model that trains on SD-VAE-style latents, including DiT and SiT, can inherit the improvement simply by training on ReaLS latents, with no architecture or training-loop changes.
  • Generated latents carry DINOv2-like semantics, so a segmentation head and a depth head can be applied directly to latents from the diffusion model, enabling training-free perception of generated images.
  • Latent-space semantic alignment and diffusion-feature alignment are complementary: combining ReaLS with REPA gives FID 23.40 at 400k steps and 18.96 at 1000k steps, compared with 33.0 for the SiT-B/2 baseline.
  • The KL weight is a key dial: the FID-versus-KL curve is U-shaped, and the chosen weight trades off the spread of the latent distribution against how much semantic alignment survives.
  • Higher reconstruction quality is not a reliable proxy for better generation: ReaLS has slightly worse rFID, PSNR, and SSIM than SD-VAE yet produces substantially better FID.

Reading between the lines

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

  • Beyond the paper: the same alignment recipe should transfer to text-to-image and video LDMs, wherever a semantic encoder can be spatially matched to the VAE's patch grid, but that transfer remains untested.
  • Beyond the paper: the semantic-consistency metric introduced here could serve as a cheap, diffusion-free screening test for candidate VAE changes, since it directly measures the semantic invariance the method relies on.
  • Beyond the paper: ReaLS inherits DINOv2's semantic coverage, so the latent space's semantic ceiling is set by that encoder; switching to a stronger or multimodal encoder could push both generation and perception further.
  • Beyond the paper: a clean causal test would train a control VAE with the same KL weight and normalization but with the alignment loss disabled, isolating the contribution of semantic alignment from the effects of the changed latent distribution.
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

3 major / 5 minor

Summary. The paper proposes ReaLS, a VAE training strategy that aligns VAE latents with DINOv2 image features (patch-level and cls-level) through an auxiliary alignment loss during VAE training. The resulting latent space is then used to train standard DiT and SiT diffusion models without modifying their architecture or training procedure. The authors report that DiT and SiT trained on ReaLS achieve about a 15% relative FID improvement over the same models trained on SD-VAE latents, and they demonstrate training-free segmentation and depth estimation by feeding generated latents through the learned alignment network. The paper also introduces a Semantic Consistency (SC) metric to quantify semantic invariance of the latent space.

Significance. If the central causal claim holds, the contribution is practically useful: it suggests a simple VAE-side modification that improves downstream latent diffusion models without touching the diffusion model, and it offers a route toward unifying generation and perception in one latent space. The paper ships code and checkpoints, uses standard evaluation metrics (FID, sFID, IS, precision, recall), and reports comparisons against established DiT/SiT baselines. However, the main attribution of the improvement to semantic alignment is not yet isolated from confounded changes in KL weighting and latent normalization, and the SC metric largely measures what the alignment loss directly enforces. The core idea is promising and the evidence is suggestive, but the manuscript currently lacks the control experiment needed to support its strongest claim.

major comments (3)
  1. [Section 4.5 and Appendix, Table 11, Eqs. (7)-(8)] The causal role of the alignment loss Lalign in the FID improvement is not isolated. The ReaLS recipe changes three factors at once relative to the SD-VAE baseline: the KL weight is set to lambda_k=2e-5 and is explicitly tuned against FID (Figure 5), the latents are rescaled by a max-min normalization to match the SD-VAE range (Eqs. (7)-(8)), and the alignment loss is added. Appendix Table 11 alone reports that changing the normalization method from std to max-min moves FID from 40 to 32, which is a large portion of the reported gains in Table 3 (e.g., SiT-B/2 from 33.0 to 27.53). Since the ablations in Tables 5-7 vary alignment features, DINOv2 model size, and alignment network depth but never remove the alignment loss, no experiment fixes the KL weight and normalization while setting lambda_a=0. The authors should train and evaluate a control VAE with the same recipe (same initialization, same KL weight, same normalization, same training budget) but with the alignment loss disabled, and report the resulting DiT/SiT FID. Without this control, the 15% improvement cannot be attributed to semantic alignment rather than to the changed latent distribution statistics and preprocessing.
  2. [Section 3.2, Algorithm 1, and Table 2] The Semantic Consistency (SC) metric is largely circular as evidence for semantic alignment. SC computes the cosine similarity between VAE latents of two random augmentations of the same image, which is exactly the quantity that the alignment loss encourages through DINOv2's augmentation-invariant features. A VAE trained with a lower KL weight can also exhibit higher latent variance and different clustering behavior even without semantic alignment, so the higher SC in Table 2 does not by itself demonstrate that the latent space is semantically structured. To support the semantic-consistency claim, the authors should report SC for the proposed control VAE without alignment loss (and with matched KL weight and normalization), and ideally compare against additional external semantic probes that are not part of the training objective.
  3. [Section 4.5 and Figure 5] The KL-weight selection procedure introduces a selection-on-FID confound. Figure 5 shows that the KL weight is chosen to minimize the FID of SiT-B/2 at 400k steps, so the comparison against the original SD-VAE baseline includes the benefit of an FID-tuned prior scale in addition to any benefit from semantic alignment. The manuscript should clarify how the SD-VAE baseline's KL weight was set and should report the FID of the control VAE (lambda_a=0) across the same KL-weight sweep. This would separate the effect of KL tuning from the effect of alignment.
minor comments (5)
  1. [Abstract and Section 4.3] The phrase '15% improvement in FID' is ambiguous because lower FID is better; please specify that the improvement is a relative reduction in FID, e.g., 'a 15% relative FID reduction.'
  2. [Table 3] The table formatting is broken for the DiT-B/2 row: the SD-VAE and ReaLS entries are merged on one line without a row separator, making the comparison difficult to read.
  3. [Section 4.4] The downstream experiments are described as 'training-free,' but they use a segmentation head from a GitHub repository and the MoGe depth head. Please clarify that only the diffusion model and VAE are frozen, while the task heads are pretrained and not fine-tuned on generated latents; the term 'training-free' should be qualified.
  4. [Algorithm 1] The SC metric does not specify which random augmentations are used, nor whether the cosine similarity is computed on raw latents or on the normalized latents used for diffusion training. Please specify these details, because normalization changes the scale of the latent vectors and therefore affects cosine similarity.
  5. [Table 4 and Section 4.3] There is inconsistent capitalization ('DIT-B/2' vs. 'DiT-B/2') and inconsistent use of dashes for unreported metrics; please standardize these entries.

Circularity Check

1 steps flagged · score 3.0 of 10

FID claim is externally benchmarked and not circular, but the semantic-consistency (SC) evidence is largely a restatement of the alignment objective.

  1. self definitional [Section 3.2 Eq. (2); Section 4.2 Algorithm 1 and Table 2]
    "Lalign = λ1Lcos(Fvae, Fdino) + λ2LsmMSE(Fvae, Fdino) (Eq. 2); "we designed a new metric based on the latent similarity after different augmentations, denoted as semantic consistency (SC)"; "SC ← CosineSimilarity(z1, z2)" (Algorithm 1)."

    The SC metric is defined as the cosine similarity between latents of two random augmentations of the same image. The alignment loss Lalign trains the VAE so that latents, after the alignment MLP, match DINOv2 features, which are themselves augmentation-invariant. Thus Lalign directly optimizes the same augmentation-invariance property that SC measures. The paper reports higher SC as evidence that ReaLS retains more semantic information, but this is a restatement of the training objective rather than an independent probe. No control VAE is trained with λa=0 while keeping the same KL weight and max–min normalization, so the SC gain cannot be attributed to semantic alignment separately from changes in latent distribution statistics.

full rationale

The central FID claim is not circular: FID is computed on 50,000 generated samples against ImageNet using an external Inception-based metric, and DiT/SiT are trained without architecture modifications, so the comparison to SD-VAE baselines is a genuine external benchmark. No self-citation chain or imported uniqueness theorem is used. The only substantial circularity I can identify is the semantic-consistency (SC) evidence: SC measures cosine similarity between latents of two augmentations, which is essentially the augmentation-invariance property that Lalign is designed to impose through DINOv2's invariance. Reporting higher SC as evidence that the latent space 'retains semantics' is therefore a restatement of the training objective. This is partial because the FID improvement is externally measured; however, the paper does not train a control VAE with the same KL weight (2e-5) and max–min normalization but λa=0, so neither SC nor FID cleanly isolates the semantic-alignment mechanism from the KL-weight and normalization choices. The KL weight and normalization method are selected against FID (Figure 5, Table 11), which is a standard tuning confound rather than a circular reduction of the FID claim itself.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim depends on several tuned loss weights and normalization choices, plus the assumption that DINOv2 semantics transfer to VAE latents. These are not derived from first principles, but none are exotic entities. No new physical or conceptual entities are posited; the alignment network is a standard MLP used as a projector.

free parameters (6)
  • KL weight lambda_k = 2e-5
    Chosen by sweeping lambda_k against SiT-B/2 FID at 400k steps (Figure 5); the paper states that KL weight significantly affects generation quality.
  • Alignment loss weight lambda_a = 1.0
    Set by hand as part of Llatent in Equation 4; no ablation reported.
  • Cosine and MSE weights lambda_1, lambda_2 = 0.9, 0.1
    Set by hand in Equation 2; no ablation reported.
  • Adversarial and perceptual weights lambda_g, lambda_p = 0.1, 1.0
    Set by hand in Equation 3 and Appendix Equation 6.
  • Latent normalization method = max-min normalization to SD-VAE range
    Appendix Table 11 shows max-min normalization outperforms std normalization (FID 32 vs 40) and is used in the reported results.
  • Alignment network depth = 2 layers
    Ablation in Table 7 shows a 2-layer MLP outperforms 1-layer and 4-layer configurations.
assumptions (4)
  • domain assumption DINOv2 features are a useful semantic prior for improving VAE latent spaces.
    Invoked in Section 3.2 as the alignment target; the paper does not prove this, but it is plausible given self-supervised feature quality.
  • domain assumption Latents can be resized to have one-to-one patch correspondence with DINOv2 patch features.
    Section 3.2 assumes images can be resized so that (H'/p', W'/p') equals (H/p, W/p) without loss.
  • standard math Standard VAE and diffusion training assumptions hold.
    Used throughout Sections 3.1 and 3.3, including reparameterization, KL to standard normal, and noise prediction.
  • domain assumption Latent normalization does not undermine the semantic alignment benefit.
    The appendix introduces normalization to the SD-VAE range; the paper assumes this preserves the benefits of ReaLS without testing otherwise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Exploring Representation-Aligned Latent Space for Better Generation." pith.science (2026). https://pith.science/paper/P6SSMKL6

@misc{pith2026250200359,
  author       = {Pith},
  title        = {Pith review of: Exploring Representation-Aligned Latent Space for Better Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P6SSMKL6}},
  note         = {Machine review of arXiv:2502.00359}
}
read the original abstract

Generative models serve as powerful tools for modeling the real world, with mainstream diffusion models, particularly those based on the latent diffusion model paradigm, achieving remarkable progress across various tasks, such as image and video synthesis. Latent diffusion models are typically trained using Variational Autoencoders (VAEs), interacting with VAE latents rather than the real samples. While this generative paradigm speeds up training and inference, the quality of the generated outputs is limited by the latents' quality. Traditional VAE latents are often seen as spatial compression in pixel space and lack explicit semantic representations, which are essential for modeling the real world. In this paper, we introduce ReaLS (Representation-Aligned Latent Space), which integrates semantic priors to improve generation performance. Extensive experiments show that fundamental DiT and SiT trained on ReaLS can achieve a 15% improvement in FID metric. Furthermore, the enhanced semantic latent space enables more perceptual downstream tasks, such as segmentation and depth estimation.

Figures

Figures reproduced from arXiv: 2502.00359 by the authors.

Figure 1
Figure 1. Representation-Aligned Latent Space (ReaLS) pre￾serves more image semantics. a) t-SNE visualization of our latent space reveals a clear clustering, with samples from the same cate￾gory closer to each other. b) Attention map of our latents shows a significant improvement in the semantic relevance among patches. et al., 2023; Huang et al., 2023; Ho et al., 2022b). Latent diffusion models (LDMs) (Rombach et al., 2022b)… view at source ↗
Figure 2
Figure 2. The training and inference pipeline of ReaLS. During VAE training, the latents of the VAE are aligned with the features of DINOv2 using an alignment network implemented via MLP. After the VAE training concludes, latent diffusion model training is performed in this latent space. In the inference phase, the latents generated by the diffusion model are converted into corresponding generated images through the VAE decod… view at source ↗
Figure 3
Figure 3. Visualization results on ImageNet 256×256, from the SiT-XL/2 + ReaLS, with cfg=4.0 [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Training-free Downstream Tasks on Latents. The diffusion model trained in the representation-aligned latent space naturally possesses stronger semantics, enabling more downstream tasks on latents. The latents generated by diffusion can obtain semantic features through …
Figure 5
Figure 5. Figure 5: Impact of KL Constraint on Latent Space and FID. As the KL weight increases from low to high, the FID initially decreases and then begins to rise again. The size of the point represents the standard deviation of the latent space. generation quality by incorporating sem…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Align & Invert: Solving Inverse Problems with Diffusion and Flow-based Models via Representation Alignment

    cs.CV 2025-11 conditional novelty 6.0 of 10

    Adding DINOv2 representation alignment to diffusion/flow inverse-problem solvers, using corrupted measurements as proxies, improves LPIPS/FID and cuts required sampling steps.

  2. Missing Fine Details in Images: Last Seen in High Frequencies

    cs.CV 2025-09 conditional novelty 6.0 of 10

    A wavelet-based VAE that trains low- and high-frequency branches separately improves image reconstruction and diffusion generation.

  3. Transition Models: Rethinking the Generative Learning Objective

    cs.LG 2025-09 conditional novelty 6.0 of 10

    TiM trains a single diffusion-type model on arbitrary time-interval transitions, achieving strong one-step and multi-step text-to-image generation with 865M parameters.

Reference graph

Works this paper leans on

25 extracted references · 1 canonical work pages · cited by 3 Pith papers

  1. [4]

    Imagenet: A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pp. 248–255. Ieee,

    2009

  2. [6]

    Ltx-video: Realtime video latent diffusion

    HaCohen, Y ., Chiprut, N., Brazowski, B., Shalem, D., Moshe, D., Richardson, E., Levin, E., Shiran, G., Zabari, N., Gordon, O., et al. Ltx-video: Realtime video latent diffusion. arXiv preprint arXiv:2501.00103,

    arXiv

  3. [8]

    Scalable adaptive computation for iterative generation

    Jabri, A., Fleet, D., and Chen, T. Scalable adaptive computation for iterative generation. arXiv preprint arXiv:2212.11972,

    arXiv

  4. [10]

    Calculate the value by sampling 10,000 samples from the ImageNet 256 × 256, encoded by V AE aligned with DINOv2-base

    The distribution of latent space changes with kl weight. Calculate the value by sampling 10,000 samples from the ImageNet 256 × 256, encoded by V AE aligned with DINOv2-base. kl weight mean std min max SD-V AE 0.29287 4.58407 -65.730 68.3175 0 1.32886 5.42394 -48.074 38.0941 1.00E-06 -0.0463 1.3918 -8.8677 7.2072 5.00E-06 0.00251 1.0678 -7.9659 9.8775 7.5...

    work page

  5. [11]

    Autoregres- sive image generation without vector quantization

    Li, T., Tian, Y ., Li, H., Deng, M., and He, K. Autoregres- sive image generation without vector quantization. arXiv preprint arXiv:2406.11838, 2024a. Li, X., Herrmann, C., Chan, K. C., Li, Y ., Sun, D., Ma, C., and Yang, M.-H. A simple approach to unifying diffusion-based conditional generation. arXiv preprint arXiv:2410.11439, 2024b. Lu, Z., Wang, Z., H...

    arXiv

  6. [12]

    S., Boffi, N

    Ma, N., Goldstein, M., Albergo, M. S., Boffi, N. M., Vanden- Eijnden, E., and Xie, S. Sit: Exploring flow and diffusion- based generative models with scalable interpolant trans- formers. arXiv preprint arXiv:2401.08740,

    arXiv

  7. [13]

    Dinov2: Learning robust visual features without supervision

    Oquab, M., Darcet, T., Moutakanni, T., V o, H., Szafraniec, M., Khalidov, V ., Fernandez, P., Haziza, D., Massa, F., El- Nouby, A., et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193,

    arXiv

  8. [14]

    Sdxl: Im- proving latent diffusion models for high-resolution image synthesis

    Podell, D., English, Z., Lacey, K., Blattmann, A., Dockhorn, T., M ¨uller, J., Penna, J., and Rombach, R. Sdxl: Im- proving latent diffusion models for high-resolution image synthesis. arXiv preprint arXiv:2307.01952,

    arXiv

Show all 25 references
  1. [16]

    High-resolution image synthesis with latent diffusion models

    Rombach, R., Blattmann, A., Lorenz, D., Esser, P., and Ommer, B. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, pp. 10684–10695, 2022a. Rombach, R., Blattmann, A., Lorenz, D....

    arXiv

  2. [17]

    Autoregressive model beats diffusion: Llama for scalable image generation

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

    arXiv

  3. [18]

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

    Tian, K., Jiang, Y ., Yuan, Z., Peng, B., and Wang, L. Visual autoregressive modeling: Scalable image generation via next-scale prediction. arXiv preprint arXiv:2404.02905,

    arXiv

  4. [19]

    Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision

    Wang, R., Xu, S., Dai, C., Xiang, J., Deng, Y ., Tong, X., and Yang, J. Moge: Unlocking accurate monocular geometry estimation for open-domain images with optimal training supervision. arXiv preprint arXiv:2410.19115, 2024a. Wang, Z., Lu, Z., Huang, D., Zhou, C., Ouyang, W., a...

    arXiv

  5. [20]

    B., Versari, L., Sohn, K., Minnen, D., Cheng, Y ., Birodkar, V ., Gupta, A., Gu, X., et al

    Yu, L., Lezama, J., Gundavarapu, N. B., Versari, L., Sohn, K., Minnen, D., Cheng, Y ., Birodkar, V ., Gupta, A., Gu, X., et al. Language model beats diffusion–tokenizer is key to visual generation. arXiv preprint arXiv:2310.05737,

    arXiv

  6. [21]

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

    Yu, S., Kwak, S., Jang, H., Jeong, J., Huang, J., Shin, J., and Xie, S. Representation alignment for generation: Training diffusion transformers is easier than you think. arXiv preprint arXiv:2410.06940,

    arXiv

  7. [22]

    Diffusion models need visual priors 10 Exploring Representation-Aligned Latent Space for Better Generation for image generation

    Yue, X., Wang, Z., Lu, Z., Sun, S., Wei, M., Ouyang, W., Bai, L., and Zhou, L. Diffusion models need visual priors 10 Exploring Representation-Aligned Latent Space for Better Generation for image generation. arXiv preprint arXiv:2410.08531,

    arXiv

  8. [23]

    Jointnet: Extending text-to-image diffusion for dense distribution modeling

    Zhang, J., Li, S., Lu, Y ., Fang, T., McKinnon, D., Tsin, Y ., Quan, L., and Yao, Y . Jointnet: Extending text-to-image diffusion for dense distribution modeling. arXiv preprint arXiv:2310.06347,

    arXiv

  9. [24]

    Fast training of diffusion models with masked transformers

    Zheng, H., Nie, W., Vahdat, A., and Anandkumar, A. Fast training of diffusion models with masked transformers. arXiv preprint arXiv:2306.09305,

    arXiv

  10. [2013]

    Applying guidance in a limited interval improves sample and distribution quality in diffusion models

    Kynk¨a¨anniemi, T., Aittala, M., Karras, T., Laine, S., Aila, T., and Lehtinen, J. Applying guidance in a limited interval improves sample and distribution quality in diffusion models. arXiv preprint arXiv:2404.07724,

    arXiv

  11. [2017]

    J., Norouzi, M., and Salimans, T

    Ho, J., Saharia, C., Chan, W., Fleet, D. J., Norouzi, M., and Salimans, T. Cascaded diffusion models for high fidelity image generation. Journal of Machine Learning Research, 23(47):1–33, 2022a. Ho, J., Salimans, T., Gritsenko, A., Chan, W., Norouzi, M., and Fleet, D. J. Video...

    arXiv

  12. [2018]

    Emu: Enhancing image generation models us- ing photogenic needles in a haystack

    Dai, X., Hou, J., Ma, C.-Y ., Tsai, S., Wang, J., Wang, R., Zhang, P., Vandenhende, S., Wang, X., Dubey, A., et al. Emu: Enhancing image generation models us- ing photogenic needles in a haystack. arXiv preprint arXiv:2309.15807,

    arXiv

  13. [2019]

    Scaling properties of diffusion models for perceptual tasks

    Ravishankar, R., Patel, Z., Rajasegaran, J., and Malik, J. Scaling properties of diffusion models for perceptual tasks. arXiv preprint arXiv:2411.08034,

    arXiv

  14. [2021]

    Tutorial on variational autoencoders

    Doersch, C. Tutorial on variational autoencoders. arXiv preprint arXiv:1606.05908,

    arXiv

  15. [2022]

    Deep compression autoen- coder for efficient high-resolution diffusion models.arXiv preprint arXiv:2410.10733,

    Chen, J., Cai, H., Chen, J., Xie, E., Yang, S., Tang, H., Li, M., Lu, Y ., and Han, S. Deep compression autoen- coder for efficient high-resolution diffusion models.arXiv preprint arXiv:2410.10733,

    arXiv

  16. [2023]

    Large scale gan training for high fidelity natural image synthesis

    Brock, A. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096,

    arXiv

  17. [2024]

    9 Exploring Representation-Aligned Latent Space for Better Generation Kingma, D. P. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114,

    arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.