REVIEW 2 major objections 5 minor 29 references
UniSpace: Unified Visual Representation and Scalable Multimodal Modeling
T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read Patch Reparameterization lets a frozen pretrained semantic ViT serve as the sole visual tokenizer for understanding, reconstruction, and generation, without a separate VAE.
desk verdict The controlled patch-embedding swap experiment gives the central claim real support; the main costs are reproducibility and thin high-resolution evidence. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the unified token $T_u = \mathrm{Concat}(T_s, \tilde{T}_r)$: the semantic stream $T_s = F_\phi(P_s(I))$ plus the compressed reconstruction stream $\tilde{T}_r = W_r F_\phi(P_r(I))$. Both streams run through the same frozen transformer blocks $F_\phi$, so the pretrained semantic pathway is untouched while a second embedding exposes the detail-rich residual pathway. The explicit channel split is what makes the representation generative-modelable: the balanced flow-matching objective weights the reconstruction component separately ($\lambda_r = 0.75$), and the compression factor keeps the reconstruction distribution easy for the prior. This mechanism turns a diagnostic observation into a reusable visual tokenizer.
What would settle it
Replace the pretrained patch embedding of SigLIP2 with a random projection, freeze the transformer blocks, train identical reconstruction probes, and measure last-layer PSNR; if the gain over the pretrained pathway is much smaller than the reported +3.70 dB (e.g., below +1 dB), the claim that the patch embedding is the bottleneck collapses.
Extended reading notes
Core claim
The paper's central claim is that pixel-level information is not lost because frozen semantic Transformer blocks lack capacity, but because the pretrained patch embedding steers the same blocks toward semantic abstraction. The evidence is a controlled swap: replacing only the patch embedding of SigLIP2 with a random projection raises last-layer PSNR by $3.70$ dB with identical frozen blocks and reconstruction probes. Patch Reparameterization turns this observation into a method: keep $T_s = F_\phi(P_s(I))$ for semantics, add $T_r = F_\phi(P_r(I))$ with $P_r$ learned for reconstruction, compress it as $\tilde{T}_r = W_r T_r$, and form the unified representation $T_u = \mathrm{Concat}(T_s, \tilde{T}_r)$. Trained with a balanced flow-matching objective ($\lambda_r = 0.75$), the resulting tokenizers reconstruct ImageNet with rFID $0.14$--$0.18$ while keeping multimodal understanding scores at or above the original encoders, and UniSpace scales the same frozen tokenizer to an 8B mixture-of-experts model for generation and editing.
Load-bearing premise
The method assumes that a second patch embedding, trained only for reconstruction, can push enough fine detail through the same frozen transformer blocks that the combined token stream decodes to near-VAE fidelity while still being easy for a generative prior to model.
Editorial extensions
If this is right
- A frozen pretrained semantic ViT can be the sole visual tokenizer of a multimodal system, eliminating the separate VAE encoder and VAE latent space used by prior unified models.
- The reconstruction gap to dedicated autoencoders closes: PR-DINOv2 reaches rFID 0.14 and PSNR 30.84 on ImageNet-1K, below RAE's 0.57 and RAEv2's 0.29 while using the smaller frozen backbone.
- Understanding does not have to be traded away: PR-SigLIP2 and PR-Qwen-ViT match or exceed their original encoders' downstream VLM accuracy when the full unified representation $T_u$ is used.
- At system scale, UniSpace (8B) reaches 4.28 on ImgEdit and 0.84 on GenEval, showing the single-space interface supports practical instruction-based editing and text-to-image generation.
- The explicit channel factorization, not just joint semantic-plus-reconstruction capability, is what makes the representation generation-friendly; an entangled MLP-merged variant fails to generate decodable latents (FID 120.9).
Reading between the lines
- If the bottleneck diagnosis holds across architectures, then any frozen semantic encoder with a residual pathway could be reparameterized the same way, so the recipe may transfer to video, audio, or multimodal encoders without retraining their backbones.
- The random-projection result (+3.70 dB) suggests deliberately detail-biased input parameterizations, such as high-frequency or multi-scale embeddings, might extract even more pixel information through the same blocks than an embedding initialized from the semantic one.
- The 95% semantic-variation measurement implies a general design rule for unified tokenizers: keep semantic and reconstruction directions explicitly separable whenever a generative prior must control both, rather than relying on learned entanglement.
- A natural scale test: applying Patch Reparameterization to larger ViTs or higher resolutions would show whether the reconstruction gain is constant or shrinks as semantic abstraction strengthens with depth.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes Patch Reparameterization, a method that turns a frozen pretrained semantic ViT into a unified visual representation supporting understanding, reconstruction, generation, and editing. The central diagnostic (Fig. 1) replaces the pretrained patch embedding of SigLIP2 with a random projection and shows that the same frozen Transformer blocks can carry pixel detail, supporting the claim that the patch embedding, not the Transformer blocks, is the reconstruction bottleneck. The method keeps the original semantic patch embedding and frozen blocks, adds a trainable reconstruction-aware patch embedding through the same blocks, compresses and concatenates the two token streams, and trains a decoder with a balanced flow-matching objective. The resulting tokenizers (PR-SigLIP2, PR-DINOv2, PR-Qwen-ViT) are evaluated on ImageNet reconstruction, LLaVA-style understanding, and ImageNet generation with matched-backbone comparisons. The same frozen tokenizer is then scaled into UniSpace, an 8B Qwen-based Mixture-of-Transformer-Experts model trained for text-to-image generation, instruction-based editing, and visual understanding with a single visual interface. System-level results are reported on ImgEdit, GEdit, GenEval, OneIG-Bench, and DPG-Bench, with qualitative comparisons against BAGEL and SenseNova-U1.
Significance. If the central claim holds, a single frozen pretrained ViT can serve as the sole visual interface for understanding, generation, and editing, removing the separate VAE pathway that dominates many current unified multimodal models. The controlled swap experiment in Sec. 3.1/Fig. 1 is a clean and falsifiable diagnostic, and the matched-backbone comparisons in Table 3 are fair: PR-DINOv2 improves rFID from 0.57 to 0.14 relative to RAE on the same DINOv2-B backbone. The explicit channel factorization and the entangled-representation diagnostic in Sec. 5.1.4 usefully demonstrate that representation-level reconstruction quality does not guarantee generative modelability, a lesson that generalizes beyond this specific system. The authors also release code and model checkpoints, and they are appropriately careful in Sec. 5.2.3 to separate controlled tokenizer-level understanding evidence from system-level validation. The main caveats are that the tokenizer-level numbers are all at 256x256 on ImageNet, the high-resolution system claims are not quantitatively tied to the 1024 training stage, and several hyperparameters are selected on validation without error bars.
major comments (2)
- [Sec. 5.1–5.2, Tables 2 and 6–11] The headline scalability claim is not yet quantitatively supported at the 1024 resolution used in the final training stages. Unified-tokenizer evaluations are all at 256x256 on ImageNet (Sec. 5.1.1–5.1.3), while the UniSpace system is trained at 256/512/1024 resolutions (Table 2), but the system-level tables (Tables 6–11) do not state the evaluation resolution, and the 1024-resolution examples in Figs. 6–9 are qualitative. Since 'scalable multimodal modeling' is a central contribution, quantitative reconstruction, editing, or generation metrics at 1024 resolution should be reported, or the claim should be explicitly scoped to the demonstrated resolutions.
- [Sec. 5.1.3, Tables 5 and 13] The 'favorable reconstruction–generation trade-off' is asserted from a single operating point rather than a matched comparison. In Table 5, PR-DINOv2 has gFID 2.10/1.87 (no-CFG/CFG) versus RAE's 1.51/1.13, alongside rFID 0.14 versus 0.57; this demonstrates a trade-off but not that the chosen point is Pareto-favorable without a comparison at a common training budget. Table 13 reports the d_r sweep only at 20 and 40 epochs (FID 6.92–15.51), far from the Table 5 numbers, and does not identify the exact configuration that produces the Table 5 results. A matched-budget comparison against RAE/RAEv2, or a gFID-versus-rFID plot across the d_r and lambda_r sweeps, would substantiate the trade-off claim.
minor comments (5)
- [Fig. 2] The word 'Reconstrution' in the figure should be 'Reconstruction'.
- [Sec. 5.1] The text says results are reported from checkpoints trained for 80 and 800 epochs, but Table 5 does not list epoch counts; please add a column or clarify which checkpoint is used.
- [Sec. 5.2] The evaluation resolution is not stated for any of the system-level benchmarks (ImgEdit, GEdit, GenEval, OneIG-Bench, DPG-Bench); since the training curriculum in Table 2 spans 256 to 1024 resolution, the test resolution should be specified for each table.
- [Sec. 4.1] The decoder is referred to as D_u in Sec. 4.1 but as D_psi in Sec. 3; the notation should be unified.
- [Table 4 and Table 5] No error bars or multiple-seed results are reported; given the small differences in Table 4 (e.g., PR-SigLIP2 64.37 versus SigLIP2 63.39), a sentence acknowledging the absence of repeated runs would be appropriate.
Circularity Check
No circularity found: the central mechanism is established by a controlled diagnostic and evaluated against external benchmarks, with no load-bearing step reducing to its own inputs.
full rationale
The paper's derivation chain is empirically self-contained rather than circular. The central claim—that frozen ViT blocks can carry pixel detail and that the original patch parameterization suppresses it—is supported by a controlled intervention in Sec. 3.1: identical frozen Transformer blocks and identical reconstruction probes, with only the patch embedding changed from pretrained P_s to random P_rand, raise last-layer PSNR from 20.96 to 24.66 dB. This is an experimental contrast, not a quantity fitted from the conclusion. The proposed method (Secs. 3.2–3.3) then trains P_r, W_r, and D_psi under reconstruction and balanced flow-matching objectives, and the claimed benefits (reconstruction fidelity, preserved understanding, and generation trade-off) are measured on external benchmarks such as ImageNet, GenEval, ImgEdit, GEdit, and DPG-Bench against independent baselines. Hyperparameters such as λ_r = 0.75 and d_r = 128 are selected via ablations reported in Tables 12–13; they are tuned constants, not predictions derived from the method. The few overlapping-author citations (LongCat-Image-Edit, UniWorld-V2) appear only as comparison baselines and are not load-bearing. No uniqueness theorem is imported from prior work, no ansatz is smuggled in via citation, and no fitted input is renamed as a prediction. Remaining gaps, such as quantitative high-resolution system-level evaluation at 1024×1024, are evidence gaps rather than circular derivation steps.
Assumptions & free parameters
free parameters (2)
- lambda_r (reconstruction objective weight) =
0.75
- d_r (reconstruction token channels) =
128
assumptions (3)
- domain assumption The frozen Transformer blocks F_phi can propagate pixel-level details when driven by a non-semantic patch embedding.
- domain assumption Explicit channel-wise separation of T_s and eTr is sufficient to make both components addressable by a generative prior.
- standard math A linear flow-matching path between Gaussian noise and T_u is a tractable generative model of the unified representation.
Cite this review
Pith. "Pith review of UniSpace: Unified Visual Representation and Scalable Multimodal Modeling." pith.science (2026). https://pith.science/paper/CV32CTOF
@misc{pith2026260808676,
author = {Pith},
title = {Pith review of: UniSpace: Unified Visual Representation and Scalable Multimodal Modeling},
year = {2026},
howpublished = {\url{https://pith.science/paper/CV32CTOF}},
note = {Machine review of arXiv:2608.08676}
}
read the original abstract
Semantic vision encoders have become a central visual interface for multimodal understanding and semantic conditioning in image generation. However, their final tokens discard fine-grained visual details, leading to poor pixel reconstruction and limiting their use in reconstruction-sensitive tasks such as image generation and editing. In this work, we ask whether understanding, generation, and editing can be modeled in a single visual representation space built from a pretrained semantic ViT. We show that the frozen Transformer blocks of a semantic ViT are not intrinsically unable to preserve visual details. Instead, the original patch parameterization drives the representation toward semantic abstraction, making fine-grained information difficult to recover from the final tokens. Based on this observation, we introduce \emph{Patch Reparameterization}, which preserves the original semantic pathway while adding a reconstruction-aware patch embedding that provides fine-grained visual information to the same frozen ViT blocks. The resulting unified representation preserves multimodal understanding while enabling high-fidelity image reconstruction and a favorable reconstruction--generation trade-off. We further scale this representation into \emph{UniSpace}, an 8B Mixture-of-Transformer-Experts model that performs understanding, generation, and editing in the same visual space without a separate VAE pathway. System-level evaluations demonstrate practical text-to-image generation and instruction-based image editing, showing that a reparameterized pretrained ViT can serve as a unified visual interface for scalable multimodal modeling.
Figures
Figures from the paper (5 more)
Reference graph
Works this paper leans on
-
[1]
Qwen3-vl technical report.arXiv preprint arXiv:2511.21631,
Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, et al. Qwen3-vl technical report.arXiv preprint arXiv:2511.21631,
-
[4]
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, 2025a. Xiaokang Chen, Zhiyu Wu, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, and C...
-
[6]
Emerging properties in unified multimodal pretraining.arXiv preprint arXiv:2505.14683,
Chaorui Deng, Deyao Zhu, Kunchang Li, Chenhui Gou, Feng Li, Zeyu Wang, Shu Zhong, Weihao Yu, Xiaonan Nie, Ziang Song, Guang Shi, and Haoqi Fan. Emerging properties in unified multimodal pretraining.arXiv preprint arXiv:2505.14683,
-
[8]
URLhttps://arxiv.org/abs/ 2506.15742. Zongjian Li, Zheyuan Liu, Qihui Zhang, Bin Lin, Feize Wu, Shenghai Yuan, Zhiyuan Yan, Yang Ye, Wangbo Yu, Yuwei Niu, et al. Uniworld-v2: Reinforce image editing with diffusion negative-aware finetuning and mllm implicit feedback.arXiv preprint arXiv:2510.16888,
-
[9]
Bin Lin, Zongjian Li, Xinhua Cheng, Yuwei Niu, Yang Ye, Xianyi He, Shenghai Yuan, Wangbo Yu, Shaodong Wang, Yunyang Ge, et al. Uniworld-v1: High-resolution semantic encoders for unified visual understanding and genera- tion.arXiv preprint arXiv:2506.03147, 2025a. Haokun Lin, Teng Wang, Yixiao Ge, Yuying Ge, Zhichao Lu, Ying Wei, Qingfu Zhang, Zhenan Sun, ...
arXiv 2024
-
[11]
Sensenova-u1: Unifying multimodal understanding and generation with neo-unify architecture
SenseNova-U1 Team. Sensenova-u1: Unifying multimodal understanding and generation with neo-unify architecture. arXiv preprint arXiv:2605.12500,
-
[12]
Improved baselines with representation autoencoders.arXiv preprint arXiv:2605.18324,
Jaskirat Singh, Boyang Zheng, Zongze Wu, Richard Zhang, Eli Shechtman, and Saining Xie. Improved baselines with representation autoencoders.arXiv preprint arXiv:2605.18324,
-
[13]
Peize Sun, Yi Jiang, Shoufa Chen, Shilong Zhang, Bingyue Peng, Ping Luo, and Zehuan Yuan. Autoregressive model beats diffusion: Llama for scalable image generation.arXiv preprint arXiv:2406.06525,
Show all 29 references
-
[14]
Unilip: Adapting clip for unified multimodal understanding, generation and editing.arXiv preprint arXiv:2507.23278,
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.arXiv preprint arXiv:2507.23278,
-
[15]
Longcat-image technical report.arXiv preprint arXiv:2512.07584,
Meituan LongCat Team, Hanghang Ma, Haoxian Tan, Jiale Huang, Junqiang Wu, Jun-Yan He, Lishuai Gao, Songlin Xiao, Xiaoming Wei, Xiaoqi Ma, et al. Longcat-image technical report.arXiv preprint arXiv:2512.07584,
-
[16]
Internvl-u: Democratizing unified multimodal models for understanding, reason- ing, generation and editing.arXiv preprint arXiv:2603.09877,
Changyao Tian, Danni Yang, Guanzhou Chen, Erfei Cui, Zhaokai Wang, Yuchen Duan, Penghao Yin, Sitao Chen, Ganlin Yang, Mingxin Liu, et al. Internvl-u: Democratizing unified multimodal models for understanding, reason- ing, generation and editing.arXiv preprint arXiv:2603.09877,
-
[17]
Siglip 2: Multilingual vision-language en- coders with improved semantic understanding, localization, and dense features.arXiv preprint arXiv:2502.14786,
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 en- coders with improved semantic understanding, localization, an...
-
[18]
Wan: Open and advanced large-scale video generative models.arXiv preprint arXiv:2503.20314,
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, Pande...
-
[19]
Ovis-u1 technical report.arXiv preprint arXiv:2506.23044,
Guo-Hua Wang, Shanshan Zhao, Xinjie Zhang, Liangfu Cao, Pengxin Zhan, Lunhao Duan, Shiyin Lu, Minghao Fu, Xiaohao Chen, Jianshan Zhao, et al. Ovis-u1 technical report.arXiv preprint arXiv:2506.23044,
-
[20]
Emu3: Next-token prediction is all you need.arXiv preprint arXiv:2409.18869,
Xinlong Wang, Xiaosong Zhang, Zhengxiong Luo, Quan Sun, Yufeng Cui, Jinsheng Wang, Fan Zhang, Yueze Wang, Zhen Li, Qiying Yu, et al. Emu3: Next-token prediction is all you need.arXiv preprint arXiv:2409.18869,
-
[21]
Qwen-image technical report.arXiv preprint arXiv:2508.02324, 2025a
Chenfei Wu, Jiahao Li, Jingren Zhou, Junyang Lin, Kaiyuan Gao, Kun Yan, Sheng-ming Yin, Shuai Bai, Xiao Xu, Yilei Chen, et al. Qwen-image technical report.arXiv preprint arXiv:2508.02324, 2025a. Chenyuan Wu, Pengfei Zheng, Ruiran Yan, Shitao Xiao, Xin Luo, Yueze Wang, Wanli Li...
2025 arXiv
-
[22]
Show-o: One single transformer to unify multimodal understanding and generation
Jinheng Xie, Weijia Mao, Zechen Bai, David Junhao Zhang, Weihao Wang, Kevin Qinghong Lin, Yuchao Gu, Zhijie Chen, Zhenheng Yang, and Mike Zheng Shou. Show-o: One single transformer to unify multimodal understanding and generation. InInternational Conference on Learning Represe...
2025
-
[23]
Qwen3 technical report.arXiv preprint arXiv:2505.09388,
An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388,
-
[24]
Towards scalable pre-training of visual tokenizers for generation.arXiv preprint arXiv:2512.13687, 2025a
Jingfeng Yao, Yuda Song, Yucong Zhou, and Xinggang Wang. Towards scalable pre-training of visual tokenizers for generation.arXiv preprint arXiv:2512.13687, 2025a. Jingfeng Yao, Bin Yang, and Xinggang Wang. Reconstruction vs. generation: Taming optimization dilemma in latent di...
-
[25]
Rep- resentation alignment for generation: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940,
Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Rep- resentation alignment for generation: Training diffusion transformers is easier than you think.arXiv preprint arXiv:2410.06940,
-
[26]
Uniflow: A unified pixel flow tokenizer for visual understanding and generation
Zhengrong Yue, Haiyu Zhang, Xiangyu Zeng, Boyu Chen, Chenting Wang, Shaobin Zhuang, Lu Dong, Yi Wang, Limin Wang, and Yali Wang. Uniflow: A unified pixel flow tokenizer for visual understanding and generation. arXiv preprint arXiv:2510.10575,
-
[27]
Qlip: Text-aligned visual tokenization unifies auto-regressive multimodal understanding and generation
Yue Zhao, Fuzhao Xue, Scott Reed, Linxi Fan, Yuke Zhu, Jan Kautz, Zhiding Yu, Philipp Kr ¨ahenb¨uhl, and De-An Huang. Qlip: Text-aligned visual tokenization unifies auto-regressive multimodal understanding and generation. arXiv preprint arXiv:2502.05178,
-
[28]
Diffusion transformers with representation autoen- coders.arXiv preprint arXiv:2510.11690,
Boyang Zheng, Nanye Ma, Shengbang Tong, and Saining Xie. Diffusion transformers with representation autoen- coders.arXiv preprint arXiv:2510.11690,
-
[29]
All variants share the same encoder, decoder, training schedule, and sampling setting, isolating the effect of the semantic–reconstruction balance in the flow-matching objective
27 UniSpace A ABLATION ONBALANCEDFLOWMATCHING We ablate the reconstruction-component objective weightλ r under the same768+128unified representation used by our final model. All variants share the same encoder, decoder, training schedule, and sampling setting, isolating the ef...
-
[2021]
Oneig-bench: Omni-dimensional nuanced evaluation for image generation.arXiv preprint arXiv:2506.07977,
Jingjing Chang, Yixiao Fang, Peng Xing, Shuhan Wu, Wei Cheng, Rui Wang, Xianfang Zeng, Gang Yu, and Hai-Bao Chen. Oneig-bench: Omni-dimensional nuanced evaluation for image generation.arXiv preprint arXiv:2506.07977,
-
[2023]
Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135,
Xiwei Hu, Rui Wang, Yixiao Fang, Bin Fu, Pei Cheng, and Gang Yu. Ella: Equip diffusion models with llm for enhanced semantic alignment.arXiv preprint arXiv:2403.05135,
-
[2024]
Yufeng Cui, Honghao Chen, Haoge Deng, Xu Huang, Xinghang Li, Jirong Liu, Yang Liu, Zhuoyan Luo, Jin- sheng Wang, Wenxuan Wang, et al. Emu3. 5: Native multimodal models are world learners.arXiv preprint arXiv:2510.26583,
-
[2025]
Z-image: An efficient image generation foundation model with single-stream diffusion trans- former.arXiv preprint arXiv:2511.22699,
Huanqia Cai, Sihan Cao, Ruoyi Du, Peng Gao, Steven Hoi, Zhaohui Hou, Shijie Huang, Dengyang Jiang, Xin Jin, Liangchen Li, et al. Z-image: An efficient image generation foundation model with single-stream diffusion trans- former.arXiv preprint arXiv:2511.22699,
-
[2026]
Open-magvit2: An open-source project toward democratizing auto-regressive visual generation.arXiv preprint arXiv:2409.04410,
Zhuoyan Luo, Fengyuan Shi, Yixiao Ge, Yujiu Yang, Limin Wang, and Ying Shan. Open-magvit2: An open-source project toward democratizing auto-regressive visual generation.arXiv preprint arXiv:2409.04410,
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.