pith. machine review for the scientific record. sign in

arxiv: 2605.07915 · v1 · submitted 2026-05-08 · 💻 cs.CV

Recognition: 2 theorem links

· Lean Theorem

What Matters for Diffusion-Friendly Latent Manifold? Prior-Aligned Autoencoders for Latent Diffusion

Zhengrong Yue , Taihang Hu , Mengting Chen , Haiyu Zhang , Zihao Pan , Tao Liu , Zikang Wang , Jinsong Lan
show 3 more authors
Xiaoyong Zhu Bo Zheng Yali Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-11 01:53 UTC · model grok-4.3

classification 💻 cs.CV
keywords latent diffusion modelsautoencoderslatent manifoldtokenizersimage generationprior alignmentImageNet
0
0 comments X

The pith

Shaping the latent manifold with explicit priors creates spaces that let diffusion models train faster and generate higher-quality images.

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

The paper examines what kind of latent space helps diffusion models work well and concludes that its internal organization matters more than how faithfully it reconstructs the original images. Controlled experiments with different tokenizers reveal three useful properties: the latent codes preserve spatial layout, nearby points stay smoothly connected, and the overall space carries meaningful high-level information. These traits line up more closely with good final images than low reconstruction error does. To build such a space directly, the authors introduce the Prior-Aligned AutoEncoder that pulls latent codes toward refined priors from visual models and adds regularization based on small perturbations.

Core claim

The central claim is that a diffusion-friendly latent manifold requires coherent spatial structure, local manifold continuity, and global manifold semantics, and that these properties can be turned into explicit training goals rather than left to emerge indirectly; the Prior-Aligned AutoEncoder achieves this by aligning latents to refined priors from visual foundation models and applying perturbation-based regularization, resulting in faster convergence and higher generation quality for latent diffusion models.

What carries the argument

The Prior-Aligned AutoEncoder (PAE), which converts the three manifold properties into direct training objectives by using refined priors from visual foundation models together with perturbation-based regularization.

If this is right

  • Tokenizers built around the three manifold properties reach performance comparable to existing strong baselines while converging up to 13 times faster under identical training conditions.
  • Generation quality on ImageNet 256x256 improves to a new state-of-the-art gFID of 1.03.
  • Reconstruction accuracy becomes a weaker predictor of downstream diffusion success than the organization of the latent manifold.
  • Explicitly shaping the latent space outperforms approaches that rely only on reconstruction or inherited pretrained representations.

Where Pith is reading between the lines

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

  • The same three properties could be added as regularizers to other latent encoders even without access to visual foundation model priors.
  • These manifold requirements may extend to latent spaces used by autoregressive or other non-diffusion generators.
  • Testing the properties at different image resolutions or on non-ImageNet datasets would show whether they are general or tied to the current training scale.

Load-bearing premise

The three manifold properties are the main cause of the observed gains in speed and quality rather than other unmeasured differences in tokenizer design or training details.

What would settle it

A controlled tokenizer variant that lacks one or more of the three properties but still reaches the same generation quality and training speed as PAE, or an experiment where enforcing the properties produces no measurable improvement in diffusion results.

Figures

Figures reproduced from arXiv: 2605.07915 by Bo Zheng, Haiyu Zhang, Jinsong Lan, Mengting Chen, Taihang Hu, Tao Liu, Xiaoyong Zhu, Yali Wang, Zhengrong Yue, Zihao Pan, Zikang Wang.

Figure 1
Figure 1. Figure 1: Prior alignment constructs a diffusion-friendly latent manifold. Left: a conceptual illus￾tration of latent space under the manifold assumption [53]. Compared with the reconstruction-oriented counterpart, the prior-aligned latent manifold is more structurally coherent, locally continuous, and semantically organized. Right: PAE yields faster convergence, better generation quality, and robust few-step sampli… view at source ↗
Figure 2
Figure 2. Figure 2: Pilot experiments on diffusion-friendly latent manifold properties. (a) Better re￾construction alone (rFID) does not guarantee better generation quality (gFID). (b–d) In contrast, improvements in instance-level structure, local manifold continuity, and global manifold semantics consistently correlate with better generation across controlled tokenizer variants. Together, these motivate latent-manifold organ… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the PAE framework. A frozen VFM provides stable representation features for the input image. DAM injects pixel detail while preserving the VFM as the dominant semantic source. The modulated representation is projected into a compact latent space for downstream diffusion. On top of this backbone, three prior-alignment objectives explicitly shape the latent manifold: SSR preserves instance-level … view at source ↗
Figure 4
Figure 4. Figure 4: Refined VFM priors provide better-matched alignment targets for PAE. Left: refined structural targets exhibit clearer patch-wise spatial correlations, yielding cleaner supervision for SSR. Right: compressed semantic targets remain well clustered in embedding space, indicating improved bottleneck matching without losing semantic organization. Decoding and Reconstruction. The deprojector Qθ maps the latent c… view at source ↗
Figure 5
Figure 5. Figure 5: Class-conditional samples by PAE with LightningDiT-XL/1 show excellent image quality. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Understanding PAE’s fidelity–learnability advantage. (a) Trade-off between recon￾struction fidelity and downstream learnability across tokenizers. ∗ denotes generative performance measured at 64 training epochs. (b) Comparison of reconstruction, latent geometry, and utilization using rFID, SSC, LPC, GSQ, and eRank. (c) Profiles of PAE built on different VFM backbones. (b) Class-conditional Generation Compa… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative Comparison. (a) Reconstruction: PAE outperforms other tokenizers in reconstructing details (e.g., thin structures, text, and faces). (b) Generation: 256 × 256 ImageNet samples from LightningDiT-XL/1 (80 epochs) demonstrating the high fidelity and coherence of PAE. Convergence Speed and Final Performance. Tab. 1 reports both short-horizon convergence and final performance. At 80 generator epochs… view at source ↗
Figure 8
Figure 8. Figure 8: Generalization and design analysis. PAE remains effective across teacher encoders and is stable under moderate changes in latent dimension and DAM depth. 1.86 gFID and 210.8 IS. This confirms that PAE improves generation by jointly shaping structure, continuity, and semantics. Impact of Refined Priors. Tab. 2(b) isolates target construction under the same prior-alignment losses. Refined VFM targets consist… view at source ↗
Figure 9
Figure 9. Figure 9: Illustration of Spatial Structure Coherence (SSC). For each image, we construct a latent-token affinity graph from the tokenizer output, perform spectral clustering on the token graph, and compare the resulting token partition with object-aware panoptic labels projected to latent resolution. Higher SSC indicates better alignment between latent token grouping and object-level spatial structure. denote objec… view at source ↗
Figure 10
Figure 10. Figure 10: Detailed pipeline of the VFM refine stage. Given an input image, a frozen representation encoder produces raw VFM features Hvfm. A lightweight projector–deprojector pair compresses these features into a compact latent space and reconstructs them with a representation reconstruction loss, producing a bottleneck-matched semantic prior. To improve spatial suitability, the raw feature is also upsampled, low-p… view at source ↗
Figure 11
Figure 11. Figure 11: Visualization of the structural refinement process. From left to right, we show the raw VFM feature, the AnyUp-upsampled feature, the low-pass normalized feature, and the final refined feature after resizing back to tokenizer resolution. The refinement suppresses noisy local variation while preserving coarse spatial structure, yielding a cleaner target for SSR. The second term improves the spatial suitabi… view at source ↗
Figure 12
Figure 12. Figure 12: Cross-tokenizer validation of manifold metrics. Relationship between the proposed latent-space metrics and downstream SiT-XL [55] generation quality (without classifier-free guidance) across a diverse set of existing autoencoders and tokenizers. To make the trend direction easy to interpret, the correlations are computed after simple coordinate normalization, so that a positive slope or Pearson coefficien… view at source ↗
Figure 13
Figure 13. Figure 13: Few-step sampling performance. Results are reported for PAE (DINOv2) with LightningDiT-XL/1 trained for 800 epochs using classifier-free guidance; FAE is evaluated un￾der the same setting. Left: gFID versus inference steps. Right: IS versus inference steps. PAE quickly approaches its full-sampling performance and matches the 250-step gFID of FAE using only 15 steps, corresponding to 16.7× fewer inference … view at source ↗
Figure 14
Figure 14. Figure 14: Additional reconstruction comparisons. PAE consistently preserves finer visual details than representative tokenizer baselines. 33 [PITH_FULL_IMAGE:figures/full_fig_p033_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Additional reconstruction comparisons. PAE consistently preserves finer visual details than representative tokenizer baselines. 34 [PITH_FULL_IMAGE:figures/full_fig_p034_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Spatial structure visualization. With prior alignment, PAE produces clearer patch-wise similarity structure and better preserves object-level spatial relations. PAE without Prior-alignment DINOv2 DINOv3 PAE with Prior-alignment SigLIP2 MAE PAE without Prior-alignment PAE with Prior-alignment [PITH_FULL_IMAGE:figures/full_fig_p035_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Global semantic organization. Prior alignment yields more compact and class-consistent latent neighborhoods, improving the semantic organization of the tokenizer manifold. 35 [PITH_FULL_IMAGE:figures/full_fig_p035_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Latent interpolation in the downstream DiT space. PAE supports smooth and semanti￾cally coherent transitions, indicating a learnable and stable latent space for diffusion. 36 [PITH_FULL_IMAGE:figures/full_fig_p036_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Tokenizer latent interpolation. PAE exhibits smooth local transitions in latent space, consistent with improved local manifold continuity [PITH_FULL_IMAGE:figures/full_fig_p037_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Tokenizer latent interpolation. Interpolation trajectories in PAE latent space illustrating local continuity. 37 [PITH_FULL_IMAGE:figures/full_fig_p037_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p038_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p038_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p038_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p039_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p039_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p039_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p040_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p040_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p040_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p041_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p041_31.png] view at source ↗
Figure 32
Figure 32. Figure 32: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p041_32.png] view at source ↗
Figure 33
Figure 33. Figure 33: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p042_33.png] view at source ↗
Figure 34
Figure 34. Figure 34: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p042_34.png] view at source ↗
Figure 35
Figure 35. Figure 35: The visualization results of LightningDiT-XL/1 + PAE (DINOv2) use CFG with w = 3.3, [PITH_FULL_IMAGE:figures/full_fig_p042_35.png] view at source ↗
read the original abstract

Tokenizers are a crucial component of latent diffusion models, as they define the latent space in which diffusion models operate. However, existing tokenizers are primarily designed to improve reconstruction fidelity or inherit pretrained representations, leaving unclear what kind of latent space is truly friendly for generative modeling. In this paper, we study this question from the perspective of latent manifold organization. By constructing controlled tokenizer variants, we identify three key properties of a diffusion-friendly latent manifold: coherent spatial structure, local manifold continuity, and global manifold semantics. We find that these properties are more consistent with downstream generation quality than reconstruction fidelity. Motivated by this finding, we propose the Prior-Aligned AutoEncoder (PAE), which explicitly shapes the latent manifold instead of leaving diffusion-friendly manifold to emerge indirectly from reconstruction or inheritance. Specifically, PAE leverages refined priors derived from VFMs and perturbation-based regularization to turn spatial structure, local continuity, and global semantics into explicit training objectives. On ImageNet 256x256, PAE improves both training efficiency and generation quality over existing tokenizers, reaching performance comparable to RAE with up to 13x faster convergence under the same training setup and achieving a new state-of-the-art gFID of 1.03. These results highlight the importance of organizing the latent manifold for latent diffusion models.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The paper claims that controlled tokenizer variants reveal three key properties of a diffusion-friendly latent manifold—coherent spatial structure, local manifold continuity, and global manifold semantics—that correlate more strongly with downstream generation quality than reconstruction fidelity. It proposes the Prior-Aligned AutoEncoder (PAE), which explicitly enforces these properties via refined VFM-derived priors and perturbation-based regularization, yielding up to 13x faster convergence and a new SOTA gFID of 1.03 on ImageNet 256x256 compared to prior tokenizers.

Significance. If the causal role of the identified manifold properties holds, the work offers a principled approach to tokenizer design for latent diffusion models that prioritizes manifold organization over indirect emergence from reconstruction objectives, with potential to improve training efficiency and sample quality across generative tasks.

major comments (2)
  1. [Controlled tokenizer variants and PAE training objectives] The section on controlled tokenizer variants and PAE objectives does not report ablations that hold perturbation regularization fixed while ablating or randomizing the VFM priors (or vice versa). This leaves open the possibility that performance gains are driven primarily by prior quality rather than the three manifold properties, undermining the claim that these properties are the primary causal drivers of improved generation quality.
  2. [ImageNet 256x256 experiments] The experiments reporting 13x faster convergence and gFID 1.03 provide no details on the number of independent runs, statistical significance, variance across seeds, or exact hyperparameter matching with baselines such as RAE. Without these, the robustness of the efficiency and SOTA claims cannot be fully assessed.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the positive assessment of the work's potential impact. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Controlled tokenizer variants and PAE training objectives] The section on controlled tokenizer variants and PAE objectives does not report ablations that hold perturbation regularization fixed while ablating or randomizing the VFM priors (or vice versa). This leaves open the possibility that performance gains are driven primarily by prior quality rather than the three manifold properties, undermining the claim that these properties are the primary causal drivers of improved generation quality.

    Authors: We appreciate this observation. Our controlled tokenizer variants were designed to probe the three manifold properties by systematically varying tokenizer components, and the PAE objectives explicitly target those properties through VFM-derived priors and perturbation regularization. However, we agree that the suggested cross-ablations (holding one component fixed while randomizing the other) would more rigorously isolate their individual contributions and strengthen the causal argument. In the revised manuscript, we will add these experiments, including results with randomized VFM priors under fixed regularization and vice versa. revision: yes

  2. Referee: [ImageNet 256x256 experiments] The experiments reporting 13x faster convergence and gFID 1.03 provide no details on the number of independent runs, statistical significance, variance across seeds, or exact hyperparameter matching with baselines such as RAE. Without these, the robustness of the efficiency and SOTA claims cannot be fully assessed.

    Authors: We acknowledge that the current manuscript lacks these experimental details. In the revision, we will report the number of independent runs performed, include variance across seeds, discuss statistical significance of the observed improvements, and explicitly confirm that all baselines (including RAE) were trained under identical hyperparameter settings and data pipelines as our PAE models. These additions will allow readers to better evaluate the reliability of the efficiency and quality gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is empirically grounded

full rationale

The paper first constructs controlled tokenizer variants to empirically identify the three manifold properties (coherent spatial structure, local continuity, global semantics) and observes their correlation with generation quality versus reconstruction fidelity. This identification step is independent of the subsequent PAE design. PAE then adopts these properties as motivation for explicit regularizers (VFM-derived priors plus perturbation regularization), with all performance claims (13x convergence, gFID 1.03) presented as experimental results on ImageNet rather than reductions by construction. No self-definitional loops, fitted parameters renamed as predictions, or load-bearing self-citations appear in the derivation chain. The overall argument remains self-contained with separate empirical support.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that the three manifold properties are the decisive factors for diffusion performance and that VFM priors plus perturbation regularization can reliably enforce them. No explicit free parameters or invented entities are described.

axioms (1)
  • domain assumption Coherent spatial structure, local manifold continuity, and global manifold semantics are the primary properties that make a latent space diffusion-friendly.
    Derived from the controlled tokenizer variant experiments described in the abstract.

pith-pipeline@v0.9.0 · 5570 in / 1257 out tokens · 42054 ms · 2026-05-11T01:53:14.580309+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

106 extracted references · 106 canonical work pages · 22 internal anchors

  1. [1]

    Latent forcing: Reordering the diffusion trajectory for pixel-space image generation

    Alan Baade, Eric Ryan Chan, Kyle Sargent, Changan Chen, Justin Johnson, Ehsan Adeli, and Li Fei-Fei. Latent forcing: Reordering the diffusion trajectory for pixel-space image generation, 2026. URLhttps://arxiv.org/abs/2602.11401

    work page arXiv 2026

  2. [2]

    Flextok: Resampling images into 1d token sequences of flexible length, 2025

    Roman Bachmann, Jesse Allardice, David Mizrahi, Enrico Fini, O˘guzhan Fatih Kar, Elmira Amirloo, Alaaeldin El-Nouby, Amir Zamir, and Afshin Dehghan. Flextok: Resampling images into 1d token sequences of flexible length, 2025. URL https://arxiv.org/abs/2502. 13967

    work page 2025

  3. [3]

    Beit: Bert pre-training of image transformers

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. Beit: Bert pre-training of image transformers. InInternational Conference on Learning Representations

    work page

  4. [4]

    VFM-VAE: Vision Foundation Models Can Be Good Tokenizers for Latent Diffusion Models

    Tianci Bi, Xiaoyi Zhang, Yan Lu, and Nanning Zheng. Vision foundation models can be good tokenizers for latent diffusion models.arXiv preprint arXiv:2510.18457, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Laminating representation autoencoders for efficient diffusion, 2026

    Ramón Calvo-González and François Fleuret. Laminating representation autoencoders for efficient diffusion, 2026. URLhttps://arxiv.org/abs/2602.04873

    work page arXiv 2026

  6. [6]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 9650–9660, 2021

    work page 2021

  7. [7]

    Dino-sae: Dino spherical autoencoder for high-fidelity image reconstruction and generation, 2026

    Hun Chang, Byunghee Cha, and Jong Chul Ye. Dino-sae: Dino spherical autoencoder for high-fidelity image reconstruction and generation, 2026. URL https://arxiv.org/abs/ 2601.22904

    work page internal anchor Pith review arXiv 2026

  8. [8]

    Videojam: Joint appearance-motion representations for enhanced motion generation in video models

    Hila Chefer, Uriel Singer, Amit Zohar, Yuval Kirstain, Adam Polyak, Yaniv Taigman, Lior Wolf, and Shelly Sheynin. Videojam: Joint appearance-motion representations for enhanced motion generation in video models. InForty-second International Conference on Machine Learning

    work page

  9. [9]

    Aligning visual foundation encoders to tokenizers for diffusion models

    Bowei Chen, Sai Bi, Hao Tan, He Zhang, Tianyuan Zhang, Zhengqi Li, Yuanjun Xiong, Jianming Zhang, and Kai Zhang. Aligntok: Aligning visual foundation encoders to tokenizers for diffusion models, 2026. URLhttps://arxiv.org/abs/2509.25162

    work page arXiv 2026

  10. [10]

    Masked autoencoders are effective tokenizers for diffusion models.ArXiv, abs/2502.03444, 2025

    Hao Chen, Yujin Han, Fangyi Chen, Xiang Li, Yidong Wang, Jindong Wang, Ze Wang, Zicheng Liu, Difan Zou, and Bhiksha Raj. Masked autoencoders are effective tokenizers for diffusion models, 2025. URLhttps://arxiv.org/abs/2502.03444

    work page arXiv 2025

  11. [11]

    Vitamin: Designing scalable vision models in the vision-language era

    Jieneng Chen, Qihang Yu, Xiaohui Shen, Alan Yuille, and Liang-Chieh Chen. Vitamin: Designing scalable vision models in the vision-language era. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12954–12966, 2024

    work page 2024

  12. [12]

    BLIP3-o: A Family of Fully Open Unified Multimodal Models-Architecture, Training and Dataset

    Jiuhai Chen, Zhiyang Xu, Xichen Pan, Yushi Hu, Can Qin, Tom Goldstein, Lifu Huang, Tianyi Zhou, Saining Xie, Silvio Savarese, et al. Blip3-o: A family of fully open unified multimodal models-architecture, training and dataset.arXiv preprint arXiv:2505.09568, 2025. 10

    work page Pith review arXiv 2025

  13. [13]

    Deep compression autoencoder for efficient high-resolution diffusion models.arXiv preprint arXiv:2410.10733, 2024

    Junyu Chen, Han Cai, Junsong Chen, Enze Xie, Shang Yang, Haotian Tang, Muyang Li, Yao Lu, and Song Han. Deep compression autoencoder for efficient high-resolution diffusion models, 2025. URLhttps://arxiv.org/abs/2410.10733

    work page arXiv 2025

  14. [14]

    Dc-ae 1.5: Accelerating diffusion model convergence with structured latent space, 2025

    Junyu Chen, Dongyun Zou, Wenkun He, Junsong Chen, Enze Xie, Song Han, and Han Cai. Dc-ae 1.5: Accelerating diffusion model convergence with structured latent space, 2025. URL https://arxiv.org/abs/2508.00413

    work page arXiv 2025

  15. [15]

    Vision Transformers Need Registers

    Timothée Darcet, Maxime Oquab, Julien Mairal, and Piotr Bojanowski. Vision transformers need registers.arXiv preprint arXiv:2309.16588, 2023

    work page internal anchor Pith review arXiv 2023

  16. [16]

    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. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    work page 2009

  17. [17]

    Repack then refine: Efficient diffusion transformer with vision foundation model, 2026

    Guanfang Dong, Luke Schultz, Negar Hassanpour, and Chao Gao. Repack then refine: Efficient diffusion transformer with vision foundation model, 2026. URL https://arxiv.org/abs/ 2512.12083

    work page internal anchor Pith review arXiv 2026

  18. [18]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  19. [19]

    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

    work page 2021

  20. [20]

    Esser, R

    Patrick Esser, Robin Rombach, and Björn Ommer. Taming transformers for high-resolution image synthesis, 2021. URLhttps://arxiv.org/abs/2012.09841

    work page arXiv 2021

  21. [21]

    Scaling rectified flow trans- formers 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 trans- formers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024

    work page 2024

  22. [22]

    The prism hypothesis: Harmonizing semantic and pixel representations via unified autoencoding

    Weichen Fan, Haiwen Diao, Quan Wang, Dahua Lin, and Ziwei Liu. The prism hypothesis: Harmonizing semantic and pixel representations via unified autoencoding, 2026. URL https: //arxiv.org/abs/2512.19693

    work page arXiv 2026

  23. [23]

    Datacomp: In search of the next generation of multimodal datasets.Advances in Neural Information Processing Systems, 36:27092–27112, 2023

    Samir Yitzhak Gadre, Gabriel Ilharco, Alex Fang, Jonathan Hayase, Georgios Smyrnis, Thao Nguyen, Ryan Marten, Mitchell Wortsman, Dhruba Ghosh, Jieyu Zhang, et al. Datacomp: In search of the next generation of multimodal datasets.Advances in Neural Information Processing Systems, 36:27092–27112, 2023

    work page 2023

  24. [24]

    One layer is enough: Adapting pretrained visual encoders for image generation.arXiv preprint arXiv:2512.07829,

    Yuan Gao, Chen Chen, Tianrong Chen, and Jiatao Gu. One layer is enough: Adapting pretrained visual encoders for image generation, 2025. URL https://arxiv.org/abs/ 2512.07829

    work page arXiv 2025

  25. [25]

    Rpiae: A representation-pivoted autoencoder enhancing both image generation and editing.arXiv preprint arXiv:2603.19206, 2026

    Yue Gong, Hongyu Li, Shanyuan Liu, Bo Cheng, Yuhang Ma, Liebucha Wu, Xiaoyu Wu, Manyuan Zhang, Dawei Leng, Yuhui Yin, and Lijun Zhang. Rpiae: A representation-pivoted autoencoder enhancing both image generation and editing, 2026. URL https://arxiv.org/ abs/2603.19206

    work page arXiv 2026

  26. [26]

    Adapting Self-Supervised Representations as a Latent Space for Efficient Generation

    Ming Gui, Johannes Schusterbauer, Timy Phan, Felix Krause, Josh Susskind, Miguel Angel Bautista, and Björn Ommer. Adapting self-supervised representations as a latent space for efficient generation, 2026. URLhttps://arxiv.org/abs/2510.14630

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  27. [27]

    Learnings from scaling visual tokenizers for reconstruction and generation.arXiv preprint arXiv:2501.09755, 2025

    Philippe Hansen-Estruch, David Yan, Ching-Yao Chung, Orr Zohar, Jialiang Wang, Tingbo Hou, Tao Xu, Sriram Vishwanath, Peter Vajda, and Xinlei Chen. Learnings from scaling visual tokenizers for reconstruction and generation.arXiv preprint arXiv:2501.09755, 2025

    work page arXiv 2025

  28. [28]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 11

    work page 2016

  29. [29]

    Masked autoencoders are scalable vision learners, 2021

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners, 2021. URL https://arxiv.org/abs/2111. 06377

    work page 2021

  30. [31]

    arXiv preprint arXiv:2602.17270 (2026)

    Jonathan Heek, Emiel Hoogeboom, Thomas Mensink, and Tim Salimans. Unified latents (ul): How to train your latents, 2026. URLhttps://arxiv.org/abs/2602.17270

    work page arXiv 2026

  31. [32]

    Query-Key Normalization for

    Alex Henry, Prudhvi Raj Dachapally, Shubham Pawar, and Yuxuan Chen. Query-key normal- ization for transformers.arXiv preprint arXiv:2010.04245, 2020

    work page arXiv 2010

  32. [33]

    Reducing the dimensionality of data with neural networks

    Geoffrey Hinton and Ruslan Salakhutdinov. Reducing the dimensionality of data with neural networks. 2006. URL10.1126/science.1127647

    work page 2006

  33. [34]

    The Platonic Representation Hypothesis

    Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. The platonic representation hypothesis.arXiv preprint arXiv:2405.07987, 2024

    work page Pith review arXiv 2024

  34. [35]

    What secrets do your manifolds hold? understanding the local geometry of gener- ative models.arXiv preprint arXiv:2408.08307, 2024

    Ahmed Imtiaz Humayun, Ibtihel Amara, Cristina Vasconcelos, Deepak Ramachandran, Can- dice Schumann, Junfeng He, Katherine Heller, Golnoosh Farnadi, Negar Rostamzadeh, and Mohammad Havaei. What secrets do your manifolds hold? understanding the local geometry of generative models.arXiv preprint arXiv:2408.08307, 2024

    work page arXiv 2024

  35. [36]

    Openclip, July 2021

    Gabriel Ilharco, Mitchell Wortsman, Ross Wightman, Cade Gordon, Nicholas Carlini, Rohan Taori, Achal Dave, Vaishaal Shankar, Hongseok Namkoong, John Miller, Hannaneh Hajishirzi, Ali Farhadi, and Ludwig Schmidt. Openclip, July 2021. Computer software. An open-source implementation of CLIP

    work page 2021

  36. [37]

    A style-based generator architecture for generative adversarial networks

    Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks, 2019. URLhttps://arxiv.org/abs/1812.04948

    work page arXiv 2019

  37. [38]

    Guiding a diffusion model with a bad version of itself, 2024

    Tero Karras, Miika Aittala, Tuomas Kynkäänniemi, Jaakko Lehtinen, Timo Aila, and Samuli Laine. Guiding a diffusion model with a bad version of itself, 2024. URL https://arxiv. org/abs/2406.02507

    work page arXiv 2024

  38. [40]

    Auto-Encoding Variational Bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2022. URL https: //arxiv.org/abs/1312.6114

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  39. [42]

    Eq-vae: Equivariance regularized la- tent 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, 2025. URL https://arxiv.org/abs/2502.09509

    work page arXiv 2025

  40. [43]

    Boosting generative image modeling via joint image-feature synthesis.arXiv preprint arXiv:2504.16064, 2025

    Theodoros Kouzelis, Efstathios Karypidis, Ioannis Kakogeorgiou, Spyros Gidaris, and Nikos Komodakis. Boosting generative image modeling via joint image-feature synthesis.arXiv preprint arXiv:2504.16064, 2025

    work page arXiv 2025

  41. [44]

    Imagenet classification with deep convolutional neural networks.Advances in neural information processing systems, 25, 2012

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks.Advances in neural information processing systems, 25, 2012

    work page 2012

  42. [45]

    Flux.https://github.com/black-forest-labs/flux, 2024

    Black Forest Labs. Flux.https://github.com/black-forest-labs/flux, 2024

    work page 2024

  43. [46]

    Repa-e: Unlocking vae for end-to-end tuning with latent diffusion transformers, 2025

    Xingjian Leng, Jaskirat Singh, Yunzhong Hou, Zhenchang Xing, Saining Xie, and Liang Zheng. Repa-e: Unlocking vae for end-to-end tuning with latent diffusion transformers.arXiv preprint arXiv:2504.10483, 2025. 12

    work page arXiv 2025

  44. [47]

    Imagefolder: Autoregressive image gen- eration with folded tokens.arXiv preprint arXiv:2410.01756,

    Xiang Li, Kai Qiu, Hao Chen, Jason Kuen, Jiuxiang Gu, Bhiksha Raj, and Zhe Lin. Image- folder: Autoregressive image generation with folded tokens.arXiv preprint arXiv:2410.01756, 2024

    work page arXiv 2024

  45. [48]

    Exploring plain vision transformer backbones for object detection

    Yanghao Li, Hanzi Mao, Ross Girshick, and Kaiming He. Exploring plain vision transformer backbones for object detection. InEuropean conference on computer vision, pages 280–296. Springer, 2022

    work page 2022

  46. [49]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014

    work page 2014

  47. [50]

    Geometric autoencoder for diffusion models,

    Hangyu Liu, Jianyong Wang, and Yutao Sun. Geometric autoencoder for diffusion models,

    work page

  48. [51]

    URLhttps://arxiv.org/abs/2603.10365

    work page arXiv

  49. [52]

    Delving into latent spectral biasing of video vaes for superior diffusability.arXiv preprint arXiv:2512.05394, 2025

    Shizhan Liu, Xinran Deng, Zhuoyi Yang, Jiayan Teng, Xiaotao Gu, and Jie Tang. Delving into latent spectral biasing of video vaes for superior diffusability, 2025. URL https://arxiv. org/abs/2512.05394

    work page arXiv 2025

  50. [53]

    Improving reconstruction of representation autoencoder, 2026

    Siyu Liu, Chujie Qin, Hubery Yin, Qixin Yan, Zheng-Peng Duan, Chen Li, Jing Lyu, Chun-Le Guo, and Chongyi Li. Improving reconstruction of representation autoencoder, 2026. URL https://arxiv.org/abs/2602.08620

    work page arXiv 2026

  51. [54]

    Deep generative models through the lens of the manifold hypothesis: A survey and new connections.arXiv preprint arXiv:2404.02954, 2024

    Gabriel Loaiza-Ganem, Brendan Leigh Ross, Rasa Hosseinzadeh, Anthony L Caterini, and Jesse C Cresswell. Deep generative models through the lens of the manifold hypothesis: A survey and new connections.arXiv preprint arXiv:2404.02954, 2024

    work page arXiv 2024

  52. [55]

    arXiv preprint arXiv:2502.20321 (2025) 9

    Chuofan Ma, Yi Jiang, Junfeng Wu, Jihan Yang, Xin Yu, Zehuan Yuan, Bingyue Peng, and Xiaojuan Qi. Unitok: A unified tokenizer for visual generation and understanding, 2025. URL https://arxiv.org/abs/2502.20321

    work page arXiv 2025

  53. [56]

    Sit: Exploring flow and diffusion-based generative models with scalable interpolant transformers

    Nanye Ma, Mark Goldstein, Michael S Albergo, Nicholas M Boffi, Eric Vanden-Eijnden, and Saining Xie. Sit: Exploring flow and diffusion-based generative models with scalable interpolant transformers. InEuropean Conference on Computer Vision, pages 23–40. Springer, 2024

    work page 2024

  54. [57]

    DINOv2: Learning Robust Visual Features without Supervision

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  55. [58]

    Boosting latent diffusion models via disentangled representation alignment, 2026

    John Page, Xuesong Niu, Kai Wu, and Kun Gai. Boosting latent diffusion models via disentangled representation alignment, 2026. URL https://arxiv.org/abs/2601.05823

    work page arXiv 2026

  56. [59]

    Semantics lead the way: Harmonizing semantic and texture modeling with asynchronous latent diffusion.arXiv preprint arXiv:2512.04926, 2025

    Yueming Pan, Ruoyu Feng, Qi Dai, Yuqi Wang, Wenfeng Lin, Mingyu Guo, Chong Luo, and Nanning Zheng. Semantics lead the way: Harmonizing semantic and texture modeling with asynchronous latent diffusion.arXiv preprint arXiv:2512.04926, 2025

    work page arXiv 2025

  57. [60]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    work page 2023

  58. [61]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  59. [62]

    Du, Zehuan Yuan, and Xinglong Wu

    Liao Qu, Huichao Zhang, Yiheng Liu, Xu Wang, Yi Jiang, Yiming Gao, Hu Ye, Daniel K. Du, Zehuan Yuan, and Xinglong Wu. Tokenflow: Unified image tokenizer for multimodal understanding and generation, 2025. URLhttps://arxiv.org/abs/2412.03069

    work page arXiv 2025

  60. [63]

    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. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021. 13

    work page 2021

  61. [64]

    When worse is better: Navigating the compression-generation tradeoff in visual tokenization,

    Vivek Ramanujan, Kushal Tirumala, Armen Aghajanyan, Luke Zettlemoyer, and Ali Farhadi. When worse is better: Navigating the compression-generation tradeoff in visual tokenization,

    work page

  62. [65]

    URLhttps://arxiv.org/abs/2412.16326

    work page arXiv

  63. [66]

    Generating diverse high-fidelity images with VQ-V AE-2.arXiv:1906.00446, 2019

    Ali Razavi, Aaron van den Oord, and Oriol Vinyals. Generating diverse high-fidelity images with vq-vae-2, 2019. URLhttps://arxiv.org/abs/1906.00446

    work page arXiv 2019

  64. [67]

    Faster r-cnn: Towards real-time object detection with region proposal networks.Advances in neural information processing systems, 28, 2015

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real-time object detection with region proposal networks.Advances in neural information processing systems, 28, 2015

    work page 2015

  65. [68]

    Published as a conference paper at ICLR

    Severi Rissanen, Markus Heinonen, and Arno Solin. Generative modelling with inverse heat dissipation, 2023. URLhttps://arxiv.org/abs/2206.13397

    work page arXiv 2023

  66. [69]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    work page 2022

  67. [70]

    LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs

    Christoph Schuhmann, Richard Vencu, Romain Beaumont, Robert Kaczmarczyk, Clayton Mullis, Aarush Katta, Theo Coombes, Jenia Jitsev, and Aran Komatsuzaki. Laion-400m: Open dataset of clip-filtered 400 million image-text pairs.arXiv preprint arXiv:2111.02114, 2021

    work page internal anchor Pith review arXiv 2021

  68. [71]

    GLU Variants Improve Transformer

    Noam Shazeer. Glu variants improve transformer.arXiv preprint arXiv:2002.05202, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  69. [72]

    Cat: Content-adaptive image tokenization, 2025

    Junhong Shen, Kushal Tirumala, Michihiro Yasunaga, Ishan Misra, Luke Zettlemoyer, Lili Yu, and Chunting Zhou. Cat: Content-adaptive image tokenization, 2025. URL https: //arxiv.org/abs/2501.03120

    work page arXiv 2025

  70. [73]

    Latent diffusion model without variational autoencoder.arXiv preprint arXiv:2510.15301, 2025

    Minglei Shi, Haolin Wang, Wenzhao Zheng, Ziyang Yuan, Xiaoshi Wu, Xintao Wang, Pengfei Wan, Jie Zhou, and Jiwen Lu. Latent diffusion model without variational autoencoder, 2025. URLhttps://arxiv.org/abs/2510.15301

    work page arXiv 2025

  71. [74]

    Oriane Siméoni, Huy V . V o, Maximilian Seitzer, Federico Baldassarre, Maxime Oquab, Cijo Jose, Vasil Khalidov, Marc Szafraniec, Seungeun Yi, Michaël Ramamonjisoa, Francisco Massa, Daniel Haziza, Luca Wehrstedt, Jianyuan Wang, Timothée Darcet, Théo Moutakanni, Leonel Sentana, Claire Roberts, Andrea Vedaldi, Jamie Tolan, John Brandt, Camille Couprie, Julie...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  72. [75]

    What matters for representation alignment: Global information or spatial structure? arXiv preprint arXiv:2512.10794, 2025

    Jaskirat Singh, Xingjian Leng, Zongze Wu, Liang Zheng, Richard Zhang, Eli Shechtman, and Saining Xie. What matters for representation alignment: Global information or spatial structure?arXiv preprint arXiv:2512.10794, 2025

    work page arXiv 2025

  73. [76]

    Improving the diffusability of autoencoders.arXiv preprint arXiv:2502.14831, 2025

    Ivan Skorokhodov, Sharath Girish, Benran Hu, Willi Menapace, Yanyu Li, Rameen Abdal, Sergey Tulyakov, and Aliaksandr Siarohin. Improving the diffusability of autoencoders, 2025. URLhttps://arxiv.org/abs/2502.14831

    work page arXiv 2025

  74. [77]

    Roformer: Enhanced transformer with rotary position embedding.Neurocomputing, 568:127063, 2024

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

    work page 2024

  75. [78]

    arXiv preprint arXiv:2507.23278 , year=

    Hao Tang, Chenwei Xie, Xiaoyi Bao, Tingyu Weng, Pandeng Li, Yun Zheng, and Liwei Wang. Unilip: Adapting clip for unified multimodal understanding, generation and editing, 2026. URLhttps://arxiv.org/abs/2507.23278

    work page arXiv 2026

  76. [79]

    MAGI-1: Autoregressive Video Generation at Scale

    Hansi Teng, Hongyu Jia, Lei Sun, Lingzhi Li, Maolin Li, Mingqiu Tang, Shuai Han, Tianning Zhang, WQ Zhang, Weifeng Luo, et al. Magi-1: Autoregressive video generation at scale. arXiv preprint arXiv:2505.13211, 2025

    work page internal anchor Pith review arXiv 2025

  77. [80]

    arXiv preprint arXiv:2601.16208 (2026),https://arxiv.org/abs/2601.16208

    Shengbang Tong, Boyang Zheng, Ziteng Wang, Bingda Tang, Nanye Ma, Ellis Brown, Jihan Yang, Rob Fergus, Yann LeCun, and Saining Xie. Scaling text-to-image diffusion transformers with representation autoencoders, 2026. URLhttps://arxiv.org/abs/2601.16208. 14

    work page arXiv 2026

  78. [81]

    SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features

    Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, et al. Siglip 2: Multilingual vision-language encoders with improved semantic understanding, localization, and dense features.arXiv preprint arXiv:2502.14786, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  79. [82]

    A tutorial on spectral clustering, 2007

    Ulrike von Luxburg. A tutorial on spectral clustering, 2007. URL https://arxiv.org/ abs/0711.0189

    work page arXiv 2007

  80. [83]

    Wan: Open and Advanced Large-Scale Video Generative Models

    Team Wan, Ang Wang, Baole Ai, Bin Wen, Chaojie Mao, Chen-Wei Xie, Di Chen, Feiwu Yu, Haiming Zhao, Jianxiao Yang, Jianyuan Zeng, Jiayu Wang, Jingfeng Zhang, Jingren Zhou, Jinkai Wang, Jixuan Chen, Kai Zhu, Kang Zhao, Keyu Yan, Lianghua Huang, Mengyang Feng, Ningyi Zhang, Pandeng Li, Pingyu Wu, Ruihang Chu, Ruili Feng, Shiwei Zhang, Siyang Sun, Tao Fang, T...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

Showing first 80 references.