Pith. sign in

REVIEW 4 major objections 6 minor 7 cited by

Echo-4o: Harnessing the Power of GPT-4o Synthetic Images for Improved Image Generation

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read GPT-4o synthetic images complement real-world data by covering fantasy and multi-reference blind spots, improving open-source generation models.

desk verdict A useful GPT-4o synthetic dataset with two new benchmarks and consistent cross-model gains; the motivating 'synthetic vs real' question is left open, but the empirical package is worth engaging. read the letter →

arxiv 2508.09987 v1 pith:AF5CLTQK submitted 2025-08-13 cs.CV cs.AIcs.CL

Junyan Ye , Dongzhi Jiang , Zihao Wang , Leqi Zhu , Zhenghao Hu , Zilong Huang , Jun He , Zhiyuan Yan
show 4 more authors
Jinghua Yu Hongsheng Li Conghui He Weijia Li
This is my paper
classification cs.CVcs.AIcs.CL
keywords syntheticdataGPT-4otext-to-imagegenerationinstructionfollowingmulti-referencefantasyimageunifiedmultimodalmodelsbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to answer why synthetic images from GPT-4o should be used at all, given that real-world images are abundant and usually higher quality. It argues that real-world datasets miss exactly the instruction types users actually give—surreal fantasy scenes and multi-reference compositions—and that synthetic data provides cleaner, more controllable supervision for text-to-image alignment. To back this, the authors build Echo-4o-Image, a 180K-example GPT-4o-generated dataset, fine-tune the Bagel model into Echo-4o, and report consistent gains over several baselines and benchmarks. They also introduce GenEval++ and Imagine-Bench to measure instruction-following and imaginative generation more accurately. The practical point is that a relatively small, deliberately targeted synthetic dataset can improve open-source unified generation models across architectures.

What carries the argument

The load-bearing object is Echo-4o-Image, a 180K-example synthetic dataset generated by GPT-4o and deliberately concentrated on two under-covered areas: fantasy/surreal imagery and multi-reference generation. It works by providing the missing long-tailed supervision that real-world collections do not contain. The supporting measurement machinery consists of GenEval++, which replaces detector/CLIP scoring with a GPT-4.1 evaluator that checks object, count, color, position, and size criteria on 280 harder prompts, and Imagine-Bench, which rates models on attribute shifts, spatiotemporal hybridization, and multi-object imaginative compositions.

What would settle it

Fine-tune the same base model on an equal-sized real-world dataset collected for fantasy and multi-reference prompts; if the gains match Echo-4o's, the content-specific synthetic-data advantage disappears. Alternatively, train on ShareGPT-4o-Image plus multi-reference examples; matching scores would show coverage, not synthetic origin, drives the effect.

Watch

Extended reading notes

Core claim

The paper's central claim is that synthetic images generated by GPT-4o complement, rather than duplicate, real-world image data by covering rare but common-in-practice instruction types and by offering clean supervision. The authors curate Echo-4o-Image with 180K image-prompt pairs focused on fantasy and multi-reference generation. Fine-tuning Bagel on this dataset yields Echo-4o, which improves over Bagel on GenEval (0.820 to 0.895), GenEval++, DPG-Bench, and OmniContext, and dominates on the new Imagine-Bench creative-generation benchmark. The same dataset, when used to fine-tune OmniGen2 and BLIP3-o, produces consistent gains on multiple metrics, which the authors read as evidence that th

Load-bearing premise

The paper assumes that the observed performance gains come from the specific fantasy and multi-reference content of Echo-4o-Image, and that real-world data cannot provide comparable supervision, without running an equal-scale real-data fine-tune.

Editorial extensions

If this is right

  • Fine-tuning any open unified multimodal generation model on Echo-4o-Image should reproduce meaningful gains in instruction-following, so the dataset functions as a general-purpose upgrade.
  • GenEval++ can replace saturated benchmarks for instruction-following, giving a harder test that separates models in the 0.8–0.9 range.
  • Imagine-Bench adds a measurable axis for creative and fantasy generation, which standard real-world benchmarks ignore.
  • The results imply that future synthetic dataset construction should emphasize coverage of rare instructions rather than scale alone.
  • Because gains appear across Bagel, OmniGen2, and BLIP3-o, the benefit is not tied to one architecture, suggesting a shared data resource for the open-source ecosystem.

Reading between the lines

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

  • A direct test the paper leaves open: fine-tune Bagel on an equivalent-sized, carefully curated real-world dataset with fantasy and multi-reference prompts; if it matches Echo-4o's gains, the 'synthetic complements real gaps' argument would need revision.
  • The authors do not ablate content categories; an ablation removing fantasy or multi-reference subsets would show whether the gains decompose by content type or simply reflect dataset diversity.
  • If GenEval++'s GPT-4.1-based evaluation is more accurate than detectors, similar MLLM-judge protocols could become standard for compositional image-generation benchmarks.
  • The transferability result suggests that a single open synthetic dataset could be shared across many base models, lowering the cost of improving each one.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces Echo-4o-Image, a 180K-scale synthetic image dataset generated by GPT-4o, and uses it to fine-tune the unified multimodal generation model Bagel, yielding Echo-4o. The authors argue that synthetic data is valuable because it covers blind spots in real-world datasets (surreal/fantasy content, multi-reference generation, clean controllable supervision). They also propose two new benchmarks, GenEval++ (a more complex instruction-following benchmark judged by GPT-4.1) and Imagine-Bench (surreal/fantasy generation), and report that Echo-4o-Image improves several base models (Bagel, OmniGen2, BLIP3-o) across GenEval, GenEval++, DPG-Bench, and OmniContext. The main controlled comparison is against ShareGPT-4o-Image, another synthetic dataset, showing larger gains for Echo-4o-Image. The paper's central motivation, that synthetic data specifically complements real-world data gaps, is not directly tested by any real-data baseline.

Significance. If the core claim holds, the paper makes a practically useful contribution: an open, transferable synthetic dataset for improving instruction-following, fantasy generation, and multi-reference image synthesis in open-source unified models, plus two benchmarks that are harder than existing ones. The transferability experiments across three architectures are a strength, as is the explicit comparison with another GPT-4o-derived dataset. However, the significance is conditional: the motivating 'why synthetic rather than real' question is only answered indirectly, and the two new benchmarks have not been validated against human judgments or an independent evaluator. The paper would be substantially stronger with a real-data control and benchmark validation.

major comments (4)
  1. [Section 5.5 and Figure 8] The central motivating claim—that synthetic data complements real-world blind spots—is not directly tested. The only dataset-level control is ShareGPT-4o-Image, which is itself synthetic and largely derived from ALLaVA real-world pairs. No equivalent-scale fine-tune of Bagel on real image-text data is reported. The observed GenEval gain (0.895 vs. 0.838 for ShareGPT-4o-Image and 0.820 for Bagel) could stem simply from adding 180K high-quality supervised pairs, regardless of whether they are synthetic or real. Please add a real-data control of comparable scale and quality, or explicitly reframe the claim as 'targeted synthetic data beats another synthetic dataset'.
  2. [Section 4.1, GenEval++] The headline results on GenEval++ rely on GPT-4.1 as the judge, a model from the same family as the GPT-4o used to generate the training data. No human validation, inter-judge agreement, or error analysis of the judge is reported. The large reported improvement (Echo-4o 0.679 vs. Bagel 0.371) is only as trustworthy as this unvalidated evaluator. Please validate the judge on a human-annotated subset, report agreement rates, and, ideally, include a second independent judge. Also report variance or confidence intervals over the 280 prompts.
  3. [Section 4.2, Imagine-Bench] The scoring protocol for Imagine-Bench is underspecified. Table 4 reports 0–10 scores, but the text does not state who or what assigns these scores (human annotators or an MLLM), what rubric is used, or how inter-annotator reliability is measured. Without this information the benchmark is not reproducible and the reported improvements (Echo-4o 7.80 vs. Bagel 6.20) cannot be independently checked. Please provide the full evaluation protocol, including prompt templates, scoring instructions, and agreement statistics.
  4. [Section 3 and dataset description] The construction of Echo-4o-Image is not documented in enough detail for a dataset-centric paper. It is not specified how the 180K prompts were generated, what filtering and curation steps were applied, how multi-reference examples were constructed, whether any benchmark prompts overlap with training data, or what licenses and consent terms apply to the generated images. Since the dataset is the primary artifact, these details are essential for evaluating contamination risk and reproducibility.
minor comments (6)
  1. [Conclusion] Typo: 'Geneval++' should be 'GenEval++'.
  2. [Table 3] Formatting: 'boldindicates' is missing a space, and several tables have incomplete captions in the preprint. Please clean the LaTeX.
  3. [Section 5.4] The text claims consistent improvements on DPG-Bench and OmniContext, but no numeric results for these benchmarks are shown in the provided manuscript. Please include a full table or a pointer to the appendix.
  4. [Figure 8] It is unclear whether the plotted points are single runs or averaged, and no error bars or significance tests are provided. State this explicitly.
  5. [Section 4.1] The rule that anime-style or disjoint-element outputs are invalid needs a concrete automatic detection procedure; otherwise the rule may be applied inconsistently across models.
  6. [Section 5] For the baseline comparisons on the new benchmarks, please provide the exact prompt templates, sampling steps, and inference settings used for each model to ensure fair comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the core transferability claim is grounded in external benchmarks.

full rationale

The paper's central empirical claim is that fine-tuning with the Echo-4o-Image dataset improves unified multimodal generation models. This claim is evaluated primarily on established external benchmarks (GenEval, DPG-Bench, OmniContext), which are not constructed from the paper's own data or evaluator choices. The proposed GenEval++ and Imagine-Bench benchmarks are additional and are used alongside external benchmarks, not in place of them. The use of GPT-4.1 as an evaluator for GenEval++ is a methodological design choice that could introduce same-lineage evaluator bias relative to GPT-4o-generated training data, but this is not a circularity by construction: the score still measures whether a generated image satisfies the instruction, and the model could fail this test. The absence of a same-scale real-data fine-tune is a limitation of the paper's motivating comparison, not a circular derivation; it does not make the observed external-benchmark gains equivalent to the dataset's content by definition. No equation, fitted parameter, or self-citation chain is shown to reduce a reported result to the paper's own inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 3 invented entities

The paper is empirical, so the ledger lists hand-chosen training hyperparameters and the domain assumptions that carry the method: that GPT-4o data is good supervision, that the GPT-4.1 judge is unbiased, that real data actually has the claimed blind spots, and that the Imagine-Bench protocol is valid. No physical constants or standard-math axioms are involved. The dataset and benchmarks are new artifacts with external handles through their releases.

free parameters (3)
  • Dataset scale (180K images) = 180,000
    Chosen by hand; no ablation on dataset size reported, so the reader cannot tell if the gains saturate or if less data suffices.
  • Fine-tuning steps = 24,000
    Chosen without a reported sweep (Section 3); performance could depend on this stopping point.
  • Learning rate = 2e-5
    Single learning rate used; no sensitivity analysis reported.
assumptions (4)
  • domain assumption GPT-4o-generated images provide reliable, clean supervision that transfers to other architectures.
    Core of the method; tested empirically on three base models but not derived or ablated by source model.
  • domain assumption GPT-4.1's checklist-based judgments in GenEval++ are accurate and unbiased measures of instruction fidelity.
    Section 4.1: no human-correlation study or error analysis of the judge is reported.
  • domain assumption Real-world datasets have blind spots (fantasy, multi-reference, long-tail instructions) that synthetic data can fill without equivalent real-data effort.
    Figure 2 and Section 1 argue this; no equivalent-scale real-data baseline is run to confirm.
  • domain assumption Imagine-Bench human ratings (or the protocol behind Table 4) are valid; the scoring procedure is not fully described in the provided text.
    Table 4 reports scores without a stated evaluation protocol or inter-rater agreement.
invented entities (3)
  • Echo-4o-Image dataset independent evidence
    purpose: Training data distilled from GPT-4o to cover fantasy, multi-reference, and instruction-following blind spots.
    Released on HuggingFace (link provided); third parties can download and use it, giving an external handle.
  • GenEval++ benchmark independent evidence
    purpose: Harder instruction-following benchmark scored by GPT-4.1 on a five-criterion checklist.
    280 prompts are published with the paper; other labs can evaluate on them, but the judge itself is a closed model, limiting full independence.
  • Imagine-Bench benchmark independent evidence
    purpose: Evaluates surreal and fantasy generation and understanding of imaginative prompts.
    Benchmark prompts are released; however the scoring protocol is not fully specified in the provided text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Echo-4o: Harnessing the Power of GPT-4o Synthetic Images for Improved Image Generation." pith.science (2026). https://pith.science/paper/AF5CLTQK

@misc{pith2026250809987,
  author       = {Pith},
  title        = {Pith review of: Echo-4o: Harnessing the Power of GPT-4o Synthetic Images for Improved Image Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AF5CLTQK}},
  note         = {Machine review of arXiv:2508.09987}
}
read the original abstract

Recently, GPT-4o has garnered significant attention for its strong performance in image generation, yet open-source models still lag behind. Several studies have explored distilling image data from GPT-4o to enhance open-source models, achieving notable progress. However, a key question remains: given that real-world image datasets already constitute a natural source of high-quality data, why should we use GPT-4o-generated synthetic data? In this work, we identify two key advantages of synthetic images. First, they can complement rare scenarios in real-world datasets, such as surreal fantasy or multi-reference image generation, which frequently occur in user queries. Second, they provide clean and controllable supervision. Real-world data often contains complex background noise and inherent misalignment between text descriptions and image content, whereas synthetic images offer pure backgrounds and long-tailed supervision signals, facilitating more accurate text-to-image alignment. Building on these insights, we introduce Echo-4o-Image, a 180K-scale synthetic dataset generated by GPT-4o, harnessing the power of synthetic image data to address blind spots in real-world coverage. Using this dataset, we fine-tune the unified multimodal generation baseline Bagel to obtain Echo-4o. In addition, we propose two new evaluation benchmarks for a more accurate and challenging assessment of image generation capabilities: GenEval++, which increases instruction complexity to mitigate score saturation, and Imagine-Bench, which focuses on evaluating both the understanding and generation of imaginative content. Echo-4o demonstrates strong performance across standard benchmarks. Moreover, applying Echo-4o-Image to other foundation models (e.g., OmniGen2, BLIP3-o) yields consistent performance gains across multiple metrics, highlighting the datasets strong transferability.

Discussion (0). Sign in to comment.

Forward citations

Cited by 7 Pith papers

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

  1. WeEdit: A Dataset, Benchmark and Glyph-Guided Framework for Text-centric Image Editing

    cs.CV 2026-03 conditional novelty 7.0 of 10

    WeEdit trains a glyph-guided, RL-optimized image editor on a 330K-pair synthetic multilingual dataset and reports open-source SOTA on its own bilingual and multilingual text-editing benchmarks.

  2. Amortized Moment Matching for Visual Generation

    cs.LG 2026-07 accept novelty 6.0 of 10

    Amortized Fréchet Distance uses neural nets to match conditional means and covariances, yielding stronger one-step visual generators than explicit FD-loss or multi-step teachers.

  3. ExpertVerse: A General-Purpose Benchmark for Expert-Level Reasoning in Knowledge-Intensive Visual Synthesis

    cs.CV 2026-07 conditional novelty 6.0 of 10

    ExpertVerse is a new benchmark and training pipeline for knowledge-intensive image generation, and its KnowThinker model with BPPO reports state-of-the-art results on reasoning-editing tests.

  4. StructGen: Disambiguating Multi-Reference Image Generation via Structured Context Modeling

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Using identifier-based structured context instead of plain-language instructions improves multi-reference image generation in semantic alignment and reference consistency.

  5. Prompt Reinjection: Alleviating Prompt Forgetting in Multimodal Diffusion Transformers for Text-to-Image Generation

    cs.CV 2026-02 unverdicted novelty 6.0 of 10

    Re-injecting shallow text features into deeper MMDiT blocks counteracts measured 'prompt forgetting' and improves instruction following in SD3, SD3.5, FLUX, and Qwen-Image without retraining.

  6. iMontage: Unified, Versatile, Highly Dynamic Many-to-many Image Generation

    cs.CV 2025-11 conditional novelty 6.0 of 10

    iMontage repurposes a pretrained video diffusion model to generate coherent yet highly dynamic image sets from arbitrary numbers of input images.

  7. Argus-Unified: Towards A Compact and Economical Unified Model for Image Understanding and Generation

    cs.CV 2026-07 conditional novelty 5.0 of 10

    A compact unified model that reuses a frozen VLM encoder and hybrid continuous/discrete tokens reaches competitive image understanding and generation with 15.6M training images and about $2,000 in compute.

Reference graph

Works this paper leans on

87 extracted references · 17 canonical work pages · cited by 7 Pith papers

  1. [1]

    Lawrence Zitnick, Devi Parikh, and Dhruv Batra

    Aishwarya Agrawal, Jiasen Lu, Stanislaw Antol, Margaret Mitchell, C. Lawrence Zitnick, Devi Parikh, and Dhruv Batra. Vqa: Visual question answering. International Journal of Computer Vision, 123: 0 4 -- 31, 2015

    2015

  2. [2]

    Sd3-medium

    Stability AI. Sd3-medium. https://stability.ai/news/stable-diffusion-3-medium, 2024

    2024

  3. [3]

    Qwen technical report, 2023

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Sinan Tan, Jianhong Tu, Peng Wang, Shijie Wang, Wei Wang, Shengguang Wu, Benfeng X...

    arXiv 2023

  4. [4]

    Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, et al. Qwen2. 5-vl technical report. arXiv preprint arXiv:2502.13923, 2025

    arXiv 2025

  5. [5]

    Instructpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 18392--18402, 2023

    2023

  6. [6]

    Allava: Harnessing gpt4v-synthesized data for a lite vision-language model, 2024 a

    Guiming Hardy Chen, Shunian Chen, Ruifei Zhang, Junying Chen, Xiangbo Wu, Zhiyi Zhang, Zhihong Chen, Jianquan Li, Xiang Wan, and Benyou Wang. Allava: Harnessing gpt4v-synthesized data for a lite vision-language model, 2024 a

    2024

  7. [7]

    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 a

    arXiv 2025

  8. [8]

    Pixart-alpha: Fast training of diffusion transformer for photorealistic text-to-image synthesis

    Junsong Chen, Jincheng Yu, Chongjian Ge, Lewei Yao, Enze Xie, Yue Wu, Zhongdao Wang, James Kwok, Ping Luo, Huchuan Lu, et al. Pixart-alpha: Fast training of diffusion transformer for photorealistic text-to-image synthesis. arXiv preprint arXiv:2310.00426, 2023 a

    arXiv 2023

Show all 87 references
  1. [9]

    Sharegpt-4o-image: Aligning multimodal models with gpt-4o-level image generation

    Junying Chen, Zhenyang Cai, Pengcheng Chen, Shunian Chen, Ke Ji, Xidong Wang, Yunjin Yang, and Benyou Wang. Sharegpt-4o-image: Aligning multimodal models with gpt-4o-level image generation. arXiv preprint arXiv:2506.18095, 2025 b

    2025 arXiv

  2. [10]

    Sharegpt4v: Improving large multi-modal models with better captions

    Lin Chen, Jisong Li, Xiaoyi Dong, Pan Zhang, Conghui He, Jiaqi Wang, Feng Zhao, and Dahua Lin. Sharegpt4v: Improving large multi-modal models with better captions. arXiv preprint arXiv:2311.12793, 2023 b

    2023 arXiv

  3. [11]

    Sharegpt4video: Improving video understanding and generation with better captions

    Lin Chen, Xilin Wei, Jinsong Li, Xiaoyi Dong, Pan Zhang, Yuhang Zang, Zehui Chen, Haodong Duan, Zhenyu Tang, Li Yuan, et al. Sharegpt4video: Improving video understanding and generation with better captions. Advances in Neural Information Processing Systems, 37: 0 19472--19495, 2024 b

    2024

  4. [12]

    Janus-pro: Unified multimodal understanding and generation with data and model scaling

    Xiaokang Chen, Zhiyu Wu, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, and Chong Ruan. Janus-pro: Unified multimodal understanding and generation with data and model scaling. arXiv preprint arXiv:2501.17811, 2025 c

    2025 arXiv

  5. [13]

    Gonzalez, Ion Stoica, and Eric P

    Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90\ https://lmsys.org/blog/2023-03-30-vicuna/, March 2023

    2023

  6. [14]

    Training verifiers to solve math word problems

    Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021

    2021 arXiv

  7. [15]

    Emerging properties in unified multimodal pretraining

    Chaorui Deng, Deyao Zhu, Kunchang Li, Chenhui Gou, Feng Li, Zeyu Wang, Shu Zhong, Weihao Yu, Xiaonan Nie, Ziang Song, et al. Emerging properties in unified multimodal pretraining. arXiv preprint arXiv:2505.14683, 2025

    2025 arXiv

  8. [16]

    Autoregressive video generation without vector quantization

    Haoge Deng, Ting Pan, Haiwen Diao, Zhengxiong Luo, Yufeng Cui, Huchuan Lu, Shiguang Shan, Yonggang Qi, and Xinlong Wang. Autoregressive video generation without vector quantization. arXiv preprint arXiv:2412.14169, 2024

    2024 arXiv

  9. [17]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv...

    2010 arXiv

  10. [18]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas M \"u ller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first International Conference on Mach...

    2024

  11. [19]

    Geneval: An object-focused framework for evaluating text-to-image alignment, 2023 a

    Dhruba Ghosh, Hanna Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text-to-image alignment, 2023 a . URL https://arxiv.org/abs/2310.11513

    2023 arXiv

  12. [20]

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

    Dhruba Ghosh, Hannaneh Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text-to-image alignment. Advances in Neural Information Processing Systems, 36: 0 52132--52152, 2023 b

    2023

  13. [21]

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

    Dhruba Ghosh, Hannaneh Hajishirzi, and Ludwig Schmidt. Geneval: An object-focused framework for evaluating text-to-image alignment. Advances in Neural Information Processing Systems, 36, 2024

    2024

  14. [22]

    Gemini 2.0 flash

    Google. Gemini 2.0 flash. https://developers.googleblog.com/en/experiment-with-gemini-20-flash-native-image-generation, 2025

    2025

  15. [23]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948, 2025 a

    2025 arXiv

  16. [24]

    Pisa: A self-augmented data engine and training strategy for 3d understanding with large models

    Zilu Guo, Hongbin Lin, Zhihao Yuan, Chaoda Zheng, Pengshuo Qiu, Dongzhi Jiang, Renrui Zhang, Chun-Mei Feng, and Zhen Li. Pisa: A self-augmented data engine and training strategy for 3d understanding with large models. arXiv preprint arXiv:2503.10529, 2025 b

    2025 arXiv

  17. [25]

    Can we generate images with cot? let's verify and reinforce image generation step by step

    Ziyu Guo, Renrui Zhang, Chengzhuo Tong, Zhizheng Zhao, Peng Gao, Hongsheng Li, and Pheng-Ann Heng. Can we generate images with cot? let's verify and reinforce image generation step by step. arXiv preprint arXiv:2501.13926, 2025 c

    2025 arXiv

  18. [26]

    Measuring mathematical problem solving with the math dataset

    Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset. arXiv preprint arXiv:2103.03874, 2021

    2021 arXiv

  19. [27]

    Ella: Equip diffusion models with llm for enhanced semantic alignment

    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

    2024 arXiv

  20. [28]

    Tifa: Accurate and interpretable text-to-image faithfulness evaluation with question answering

    Yushi Hu, Benlin Liu, Jungo Kasai, Yizhong Wang, Mari Ostendorf, Ranjay Krishna, and Noah A Smith. Tifa: Accurate and interpretable text-to-image faithfulness evaluation with question answering. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages ...

    2023

  21. [29]

    T2i-compbench++: An enhanced and comprehensive benchmark for compositional text-to-image generation, 2025

    Kaiyi Huang, Chengqi Duan, Kaiyue Sun, Enze Xie, Zhenguo Li, and Xihui Liu. T2i-compbench++: An enhanced and comprehensive benchmark for compositional text-to-image generation, 2025. URL https://arxiv.org/abs/2307.06350

    2025 arXiv

  22. [30]

    Comat: Aligning text-to-image diffusion model with image-to-text concept matching

    Dongzhi Jiang, Guanglu Song, Xiaoshi Wu, Renrui Zhang, Dazhong Shen, Zhuofan Zong, Yu Liu, and Hongsheng Li. Comat: Aligning text-to-image diffusion model with image-to-text concept matching. arXiv preprint arXiv:2404.03653, 2024 a

    2024 arXiv

  23. [31]

    Mmsearch: Benchmarking the potential of large models as multi-modal search engines

    Dongzhi Jiang, Renrui Zhang, Ziyu Guo, Yanmin Wu, Jiayi Lei, Pengshuo Qiu, Pan Lu, Zehui Chen, Chaoyou Fu, Guanglu Song, et al. Mmsearch: Benchmarking the potential of large models as multi-modal search engines. arXiv preprint arXiv:2409.12959, 2024 b

    2024 arXiv

  24. [32]

    T2i-r1: Reinforcing image generation with collaborative semantic-level and token-level cot

    Dongzhi Jiang, Ziyu Guo, Renrui Zhang, Zhuofan Zong, Hao Li, Le Zhuo, Shilin Yan, Pheng-Ann Heng, and Hongsheng Li. T2i-r1: Reinforcing image generation with collaborative semantic-level and token-level cot. arXiv preprint arXiv:2505.00703, 2025 a

    2025 arXiv

  25. [33]

    Mme-cot: Benchmarking chain-of-thought in large multimodal models for reasoning quality, robustness, and efficiency

    Dongzhi Jiang, Renrui Zhang, Ziyu Guo, Yanwei Li, Yu Qi, Xinyan Chen, Liuhui Wang, Jianhan Jin, Claire Guo, Shen Yan, et al. Mme-cot: Benchmarking chain-of-thought in large multimodal models for reasoning quality, robustness, and efficiency. arXiv preprint arXiv:2502.09621, 2025 b

    2025 arXiv

  26. [34]

    Auto-encoding variational bayes

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  27. [35]

    Viescore: Towards explainable metrics for conditional image synthesis evaluation, 2023

    Max Ku, Dongfu Jiang, Cong Wei, Xiang Yue, and Wenhu Chen. Viescore: Towards explainable metrics for conditional image synthesis evaluation, 2023

    2023

  28. [36]

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

    2024

  29. [37]

    Genai-bench: Evaluating and improving compositional text-to-visual generation, 2024 a

    Baiqi Li, Zhiqiu Lin, Deepak Pathak, Jiayao Li, Yixin Fei, Kewen Wu, Tiffany Ling, Xide Xia, Pengchuan Zhang, Graham Neubig, and Deva Ramanan. Genai-bench: Evaluating and improving compositional text-to-visual generation, 2024 a . URL https://arxiv.org/abs/2406.13743

    2024 arXiv

  30. [38]

    Crossviewdiff: A cross-view diffusion model for satellite-to-street view synthesis

    Weijia Li, Jun He, Junyan Ye, Huaping Zhong, Zhimeng Zheng, Zilong Huang, Dahua Lin, and Conghui He. Crossviewdiff: A cross-view diffusion model for satellite-to-street view synthesis. arXiv preprint arXiv:2408.14765, 2024 b

    2024 arXiv

  31. [39]

    Hunyuan-dit: A powerful multi-resolution diffusion transformer with fine-grained chinese understanding

    Zhimin Li, Jianwei Zhang, Qin Lin, Jiangfeng Xiong, Yanxin Long, Xinchi Deng, Yingfang Zhang, Xingchao Liu, Minbin Huang, Zedong Xiao, et al. Hunyuan-dit: A powerful multi-resolution diffusion transformer with fine-grained chinese understanding. arXiv preprint arXiv:2405.08748, 2024 c

    2024 arXiv

  32. [40]

    Uniworld: High-resolution semantic encoders for unified visual understanding and generation

    Bin Lin, Zongjian Li, Xinhua Cheng, Yuwei Niu, Yang Ye, Xianyi He, Shenghai Yuan, Wangbo Yu, Shaodong Wang, Yunyang Ge, et al. Uniworld: High-resolution semantic encoders for unified visual understanding and generation. arXiv preprint arXiv:2506.03147, 2025

    2025 arXiv

  33. [41]

    Microsoft coco: Common objects in context

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

    2014

  34. [42]

    Evaluating text-to-visual generation with image-to-text generation, 2024

    Zhiqiu Lin, Deepak Pathak, Baiqi Li, Jiayao Li, Xide Xia, Graham Neubig, Pengchuan Zhang, and Deva Ramanan. Evaluating text-to-visual generation with image-to-text generation, 2024. URL https://arxiv.org/abs/2404.01291

    2024 arXiv

  35. [43]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. In NeurIPS, 2023 a

    2023

  36. [44]

    Visual instruction tuning

    Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. Visual instruction tuning. Advances in neural information processing systems, 36, 2024

    2024

  37. [45]

    Mmbench: Is your multi-modal model an all-around player? arXiv preprint arXiv:2307.06281, 2023 b

    Yuan Liu, Haodong Duan, Yuanhan Zhang, Bo Li, Songyang Zhang, Wangbo Zhao, Yike Yuan, Jiaqi Wang, Conghui He, Ziwei Liu, et al. Mmbench: Is your multi-modal model an all-around player? arXiv preprint arXiv:2307.06281, 2023 b

    2023 arXiv

  38. [46]

    The llama 3 herd of models, 2024

    Meta. The llama 3 herd of models, 2024. URL https://arxiv.org/abs/2407.21783

    2024 arXiv

  39. [47]

    OpenAI. Chatgpt. https://chat.openai.com, 2023 a

    2023

  40. [48]

    Gpt-4 technical report

    OpenAI. Gpt-4 technical report. ArXiv, abs/2303.08774, 2023 b

    2023 arXiv

  41. [49]

    GPT-4V(ision) system card, 2023 c

    OpenAI. GPT-4V(ision) system card, 2023 c . URL https://openai.com/research/gpt-4v-system-card

    2023

  42. [50]

    Dall·e 3

    OpenAI. Dall·e 3. https://openai.com/index/dall-e-3, 2024

    2024

  43. [51]

    OpenAI. Gpt-4-1. https://openai.com/index/gpt-4-1, 2025 a

    2025

  44. [52]

    OpenAI. Gpt-4o. https://openai.com/index/introducing-4o-image-generation, 2025 b

    2025

  45. [53]

    Transfer between modalities with metaqueries

    Xichen Pan, Satya Narayan Shukla, Aashu Singh, Zhuokai Zhao, Shlok Kumar Mishra, Jialiang Wang, Zhiyang Xu, Jiuhai Chen, Kunpeng Li, Felix Juefei-Xu, et al. Transfer between modalities with metaqueries. arXiv preprint arXiv:2504.06256, 2025

    2025 arXiv

  46. [54]

    Sdxl: Improving latent diffusion models for high-resolution image synthesis

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

    2023 arXiv

  47. [55]

    Tokenflow: Unified image tokenizer for multimodal understanding and generation

    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. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages ...

    2025

  48. [56]

    Hierarchical text-conditional image generation with clip latents

    Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, and Mark Chen. Hierarchical text-conditional image generation with clip latents. arXiv preprint arXiv:2204.06125, 1 0 (2): 0 3, 2022

    2022 arXiv

  49. [57]

    Gpqa: A graduate-level google-proof q&a benchmark

    David Rein, Betty Li Hou, Asa Cooper Stickland, Jackson Petty, Richard Yuanzhe Pang, Julien Dirani, Julian Michael, and Samuel R Bowman. Gpqa: A graduate-level google-proof q&a benchmark. In First Conference on Language Modeling, 2024

    2024

  50. [58]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684--10695, 2022 a

    2022

  51. [59]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Bj \"o rn Ommer. High-resolution image synthesis with latent diffusion models. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684--10695, 2022 b

    2022

  52. [60]

    Journeydb: A benchmark for generative image understanding

    Keqiang Sun, Junting Pan, Yuying Ge, Hao Li, Haodong Duan, Xiaoshi Wu, Renrui Zhang, Aojun Zhou, Zipeng Qin, Yi Wang, et al. Journeydb: A benchmark for generative image understanding. Advances in Neural Information Processing Systems, 36, 2024

    2024

  53. [61]

    Delving into rl for image generation with cot: A study on dpo vs

    Chengzhuo Tong, Ziyu Guo, Renrui Zhang, Wenyu Shan, Xinyu Wei, Zhenghao Xing, Hongsheng Li, and Pheng-Ann Heng. Delving into rl for image generation with cot: A study on dpo vs. grpo. arXiv preprint arXiv:2505.17017, 2025

    2025 arXiv

  54. [62]

    Llama: Open and efficient foundation language models

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth \'e e Lacroix, Baptiste Rozi \`e re, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  55. [63]

    Emu3: Next-token prediction is all you need

    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, 2024

    2024 arXiv

  56. [64]

    Gpt-image-edit-1.5 m: A million-scale, gpt-generated image dataset

    Yuhan Wang, Siwei Yang, Bingchen Zhao, Letian Zhang, Qing Liu, Yuyin Zhou, and Cihang Xie. Gpt-image-edit-1.5 m: A million-scale, gpt-generated image dataset. arXiv preprint arXiv:2507.21033, 2025

    2025 arXiv

  57. [65]

    Tiif-bench: How does your t2i model follow your instructions? arXiv preprint arXiv:2506.02161, 2025

    Xinyu Wei, Jinrui Zhang, Zeqing Wang, Hongyang Wei, Zhen Guo, and Lei Zhang. Tiif-bench: How does your t2i model follow your instructions? arXiv preprint arXiv:2506.02161, 2025

    2025 arXiv

  58. [66]

    Janus: Decoupling visual encoding for unified multimodal understanding and generation

    Chengyue Wu, Xiaokang Chen, Zhiyu Wu, Yiyang Ma, Xingchao Liu, Zizheng Pan, Wen Liu, Zhenda Xie, Xingkai Yu, Chong Ruan, et al. Janus: Decoupling visual encoding for unified multimodal understanding and generation. In Proceedings of the Computer Vision and Pattern Recognition ...

    2025

  59. [67]

    Omnigen2: Exploration to advanced multimodal generation

    Chenyuan Wu, Pengfei Zheng, Ruiran Yan, Shitao Xiao, Xin Luo, Yueze Wang, Wanli Li, Xiyan Jiang, Yexin Liu, Junjie Zhou, et al. Omnigen2: Exploration to advanced multimodal generation. arXiv preprint arXiv:2506.18871, 2025 b

    2025 arXiv

  60. [68]

    Less-to-more generalization: Unlocking more controllability by in-context generation

    Shaojin Wu, Mengqi Huang, Wenxu Wu, Yufeng Cheng, Fei Ding, and Qian He. Less-to-more generalization: Unlocking more controllability by in-context generation. arXiv preprint arXiv:2504.02160, 2025 c

    2025 arXiv

  61. [69]

    Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis, 2023

    Xiaoshi Wu, Yiming Hao, Keqiang Sun, Yixiong Chen, Feng Zhu, Rui Zhao, and Hongsheng Li. Human preference score v2: A solid benchmark for evaluating human preferences of text-to-image synthesis, 2023. URL https://arxiv.org/abs/2306.09341

    2023 arXiv

  62. [70]

    Omnigen: Unified image generation

    Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Chaofan Li, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 13294--13304, 2025

    2025

  63. [71]

    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. arXiv preprint arXiv:2408.12528, 2024

    2024 arXiv

  64. [72]

    Visionreward: Fine-grained multi-dimensional human preference learning for image and video generation, 2025

    Jiazheng Xu, Yu Huang, Jiale Cheng, Yuanming Yang, Jiajun Xu, Yuan Wang, Wenbo Duan, Shen Yang, Qunlin Jin, Shurun Li, Jiayan Teng, Zhuoyi Yang, Wendi Zheng, Xiao Liu, Ming Ding, Xiaohan Zhang, Xiaotao Gu, Shiyu Huang, Minlie Huang, Jie Tang, and Yuxiao Dong. Visionreward: Fin...

    2025 arXiv

  65. [73]

    Pointllm: Empowering large language models to understand point clouds

    Runsen Xu, Xiaolong Wang, Tai Wang, Yilun Chen, Jiangmiao Pang, and Dahua Lin. Pointllm: Empowering large language models to understand point clouds. arXiv preprint arXiv:2308.16911, 2023

    2023 arXiv

  66. [74]

    Gpt-imgeval: A comprehensive benchmark for diagnosing gpt4o in image generation

    Zhiyuan Yan, Junyan Ye, Weijia Li, Zilong Huang, Shenghai Yuan, Xiangyang He, Kaiqing Lin, Jun He, Conghui He, and Li Yuan. Gpt-imgeval: A comprehensive benchmark for diagnosing gpt4o in image generation. arXiv preprint arXiv:2504.02782, 2025

    2025 arXiv

  67. [75]

    The fabrication of reality and fantasy: Scene generation with llm-assisted prompt interpretation

    Yi Yao, Chan-Feng Hsu, Jhe-Hao Lin, Hongxia Xie, Terence Lin, Yi-Ning Huang, Hong-Han Shuai, and Wen-Huang Cheng. The fabrication of reality and fantasy: Scene generation with llm-assisted prompt interpretation. In European Conference on Computer Vision, pages 422--438. Springer, 2024

    2024

  68. [76]

    Skydiffusion: Street-to-satellite image synthesis with diffusion models and bev paradigm

    Junyan Ye, Jun He, Weijia Li, Zhutao Lv, Jinhua Yu, Haote Yang, and Conghui He. Skydiffusion: Street-to-satellite image synthesis with diffusion models and bev paradigm. arXiv e-prints, pages arXiv--2408, 2024 a

    2024

  69. [77]

    Loki: A comprehensive synthetic data detection benchmark using large multimodal models

    Junyan Ye, Baichuan Zhou, Zilong Huang, Junan Zhang, Tianyi Bai, Hengrui Kang, Jun He, Honglin Lin, Zihao Wang, Tong Wu, et al. Loki: A comprehensive synthetic data detection benchmark using large multimodal models. arXiv preprint arXiv:2410.09732, 2024 b

    2024 arXiv

  70. [78]

    Sigmoid loss for language image pre-training

    Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, and Lucas Beyer. Sigmoid loss for language image pre-training. In Proceedings of the IEEE/CVF international conference on computer vision, pages 11975--11986, 2023

    2023

  71. [79]

    Video-llama: An instruction-tuned audio-visual language model for video understanding

    Hang Zhang, Xin Li, and Lidong Bing. Video-llama: An instruction-tuned audio-visual language model for video understanding. arXiv preprint arXiv:2306.02858, 2023 a

    2023 arXiv

  72. [80]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 3836--3847, October 2023 b

    2023

  73. [81]

    Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In European Conference on Computer Vision, pages 169--186

    Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Yu Qiao, et al. Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In European Conference on Computer Vision, pages 169--186. Spri...

    2024

  74. [82]

    Mavis: Mathematical visual instruction tuning with an automatic data engine, 2024 b

    Renrui Zhang, Xinyu Wei, Dongzhi Jiang, Ziyu Guo, Shicheng Li, Yichi Zhang, Chengzhuo Tong, Jiaming Liu, Aojun Zhou, Bin Wei, Shanghang Zhang, Peng Gao, Chunyuan Li, and Hongsheng Li. Mavis: Mathematical visual instruction tuning with an automatic data engine, 2024 b . URL htt...

    2024 arXiv

  75. [83]

    Transfusion: Predict the next token and diffuse images with one multi-modal model

    Chunting Zhou, Lili Yu, Arun Babu, Kushal Tirumala, Michihiro Yasunaga, Leonid Shamis, Jacob Kahn, Xuezhe Ma, Luke Zettlemoyer, and Omer Levy. Transfusion: Predict the next token and diffuse images with one multi-modal model. arXiv preprint arXiv:2408.11039, 2024

    2024 arXiv

  76. [84]

    Minigpt-4: Enhancing vision-language understanding with advanced large language models

    Deyao Zhu, Jun Chen, Xiaoqian Shen, Xiang Li, and Mohamed Elhoseiny. Minigpt-4: Enhancing vision-language understanding with advanced large language models. arXiv preprint arXiv:2304.10592, 2023

    2023 arXiv

  77. [85]

    Lumina-next: Making lumina-t2x stronger and faster with next-dit

    Le Zhuo, Ruoyi Du, Han Xiao, Yangguang Li, Dongyang Liu, Rongjie Huang, Wenze Liu, Xiangyang Zhu, Fu-Yun Wang, Zhanyu Ma, et al. Lumina-next: Making lumina-t2x stronger and faster with next-dit. Advances in Neural Information Processing Systems, 37: 0 131278--131315, 2024

    2024

  78. [86]

    Easyref: Omni-generalized group image reference for diffusion models via multimodal llm

    Zhuofan Zong, Dongzhi Jiang, Bingqi Ma, Guanglu Song, Hao Shao, Dazhong Shen, Yu Liu, and Hongsheng Li. Easyref: Omni-generalized group image reference for diffusion models via multimodal llm. arXiv preprint arXiv:2412.09618, 2024 a

    2024 arXiv

  79. [87]

    Mova: Adapting mixture of vision experts to multimodal context

    Zhuofan Zong, Bingqi Ma, Dazhong Shen, Guanglu Song, Hao Shao, Dongzhi Jiang, Hongsheng Li, and Yu Liu. Mova: Adapting mixture of vision experts to multimodal context. arXiv preprint arXiv:2404.13046, 2024 b

    2024 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.