Pith. sign in

REVIEW 63 references

OmniStyle: Filtering High Quality Style Transfer Data at Scale

T0 review · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A new million-triplet dataset and a diffusion transformer model that performs text-guided and image-guided style transfer, with a filtering pipeline used to curate high-quality training examples.

arxiv 2505.14028 v1 pith:M5BMW67G submitted 2025-05-20 cs.CV

Ye Wang , Ruiqi Liu , Jiang Lin , Fei Liu , Zili Yi , Yilin Wang , Rui Ma This is my paper · ORCID
classification cs.CV
keywords styletransferframeworkhigh-qualityomnistyleomnistyle-1mqualitydataset
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

OmniStyle-1M is built in two stages. First, the authors use ChatGPT to write prompts and FLUX to generate 2,000 synthetic content images across 20 categories. They select 1,000 style images from the Style30K dataset, then run six existing style transfer models on each content-style pair to create over one million candidate triplets. A filtering pipeline called OmniFilter scores each triplet on three criteria: how much content survives, how well the style matches, and how aesthetically appealing the result is. It keeps the best candidate for each pair, yielding a 150K-triplet high-quality subset.

The OmniStyle model starts from the FLUX diffusion transformer and fine-tunes it on this filtered data. It can take either a text instruction or a reference style image as guidance. The authors report that it beats several style transfer and image editing baselines on their evaluation metrics and in a user study.

The main concerns are that the evaluation metric is the same OmniFilter score used to select training data, the paper does not say whether the benchmark images are held out from training, and neither code nor the dataset is released. The dataset itself could be a useful resource if it becomes public.

Extended reading notes

Core claim

OmniStyle-1M is the first million-level paired style transfer dataset, with over one million content-style-stylized image triplets across 1,000 style categories, and training the OmniStyle diffusion transformer on the OmniFilter-filtered subset yields state-of-the-art performance in both instruction-guided and image-guided style transfer (Abstract; Tables 2 and 3).

Load-bearing premise

The evaluation benchmark in Section 5 uses 20 content images and 100 style images, but the paper never states that these images are excluded from the OmniStyle-1M training set. If the benchmark content or style images appear during training, the reported SOTA scores in Tables 2 and 3 are inflated and the comparison is invalid. The paper also does not report any independent test split for the dataset.

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.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim relies on a handful of hand-chosen weights and several domain assumptions about the validity of CLIP/DINOv2/MLLM scorers for stylized images. No model or metric is introduced from first principles; the filtering and evaluation rest on off-the-shelf pretrained representations. The retention threshold for the 150K subset is unreported.

free parameters (3)
  • Alpha in content preservation score = 0.5
    Eq. (1) weights semantic and structural scores equally; chosen without sensitivity analysis.
  • Weights in OmniFilter total score = a=0.2, b=0.6, c=0.2
    Eq. (2) gives style 60% weight; chosen by hand, no ablation for weight choices.
  • Retention threshold for 150K subset = unspecified
    Best-of-six among ~200K pairs leaves ~150K triplets; the score threshold or rule for dropping 50K pairs is not given.
assumptions (4)
  • domain assumption CLIP image-text similarity and DINOv2 embedding similarity are valid measures of content preservation for stylized images.
    Eq. (1) combines them into Cscore without validation against human judgments on stylized outputs.
  • domain assumption A contrastively fine-tuned CLIP image encoder on Style30K labels measures style consistency between arbitrary style references and stylized outputs.
    Section 3.3 uses this scorer for both filtering and evaluation; generalization to mixed model outputs is assumed.
  • domain assumption An InternVL2 aesthetic scorer trained on AVA and BAID transfers to synthetic style-transfer images.
    Section 3.3 fine-tunes on these datasets and applies to FLUX-generated stylized images without domain validation.
  • domain assumption The six style transfer models used to generate candidates provide a sufficiently unbiased sample of successful stylizations.
    Section 3.2 assumes best-of-six approximates a good target; systematic biases shared by all six models become dataset biases.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OmniStyle: Filtering High Quality Style Transfer Data at Scale." pith.science (2026). https://pith.science/paper/M5BMW67G

@misc{pith2026250514028,
  author       = {Pith},
  title        = {Pith review of: OmniStyle: Filtering High Quality Style Transfer Data at Scale},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M5BMW67G}},
  note         = {Machine review of arXiv:2505.14028}
}
read the original abstract

In this paper, we introduce OmniStyle-1M, a large-scale paired style transfer dataset comprising over one million content-style-stylized image triplets across 1,000 diverse style categories, each enhanced with textual descriptions and instruction prompts. We show that OmniStyle-1M can not only enable efficient and scalable of style transfer models through supervised training but also facilitate precise control over target stylization. Especially, to ensure the quality of the dataset, we introduce OmniFilter, a comprehensive style transfer quality assessment framework, which filters high-quality triplets based on content preservation, style consistency, and aesthetic appeal. Building upon this foundation, we propose OmniStyle, a framework based on the Diffusion Transformer (DiT) architecture designed for high-quality and efficient style transfer. This framework supports both instruction-guided and image-guided style transfer, generating high resolution outputs with exceptional detail. Extensive qualitative and quantitative evaluations demonstrate OmniStyle's superior performance compared to existing approaches, highlighting its efficiency and versatility. OmniStyle-1M and its accompanying methodologies provide a significant contribution to advancing high-quality style transfer, offering a valuable resource for the research community.

Figures

Figures reproduced from arXiv: 2505.14028 by the authors.

Figure 1
Figure 1. OmniStyle enables high-quality (a) instruction-guided style transfer and (b) reference image-guided style transfer, covering a [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of OmniStyle-1M. (a) The inner ring represents the eight primary categories, while the outer ring corresponds to [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Overview of our dataset creation and filtering pipeline. (a) Content Image Generation: We utilize ChatGPT to automatically [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: The architecture of OmniStyle. combined score is calculated as follows: Score = a · Cscore + b · Sscore + c · Ascore, (2) where a = 0.2, b = 0.6, and c = 0.2. The output with the highest total score among the six was selected as the final sample. This filtering yielded…
Figure 5
Figure 5. Figure 5: Qualitative comparison with other state-of-the-art methods for the instruction-guided style transfer task. For clarity, the style [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Qualitative comparison with other state-of-the-art methods for the image-guided style transfer task. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Examples of content images across 20 categories. Each image is randomly selected from its respective category, showcasing the [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: A subset of style images in OmniStyle Dataset. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Examples of Triplet from the OmniStyle-1M dataset. From left to right, each set of images is arranged as: content image, style [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Frameworks for content preservation evaluation and style consistency evaluation. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Frameworks for aesthetic appeal evaluation. [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: The similarity evaluation results for different style im [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: Qualitative ablation study results for the core components of OmniFilter. The yellow highlights indicate the highest values, [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]
Figure 14
Figure 14. Figure 14: The data filtering and model training pipeline for LoRA-based style transfer task. [PITH_FULL_IMAGE:figures/full_fig_p015_14.png]
Figure 15
Figure 15. Figure 15: Additional image-guided style transfer results of OmniStyle. Left: content image, middle: style reference, right: stylized output. [PITH_FULL_IMAGE:figures/full_fig_p016_15.png]
Figure 16
Figure 16. Figure 16: Additional image-guided style transfer results of OmniStyle. Left: content image, middle: style reference, right: stylized output. [PITH_FULL_IMAGE:figures/full_fig_p017_16.png]
Figure 17
Figure 17. Figure 17: Additional image-guided style transfer results of OmniStyle. Left: content image, middle: style reference, right: stylized output. [PITH_FULL_IMAGE:figures/full_fig_p018_17.png]
Figure 18
Figure 18. Figure 18: Additional image-guided style transfer results of OmniStyle. Left: content image, middle: style reference, right: stylized output. [PITH_FULL_IMAGE:figures/full_fig_p019_18.png]
Figure 19
Figure 19. Figure 19: Additional image-guided style transfer results of OmniStyle. Left: content image, middle: style reference, right: stylized output. [PITH_FULL_IMAGE:figures/full_fig_p020_19.png]
Figure 20
Figure 20. Figure 20: Additional instruction-guided style transfer results of OmniStyle. The left column shows the content images, the middle column [PITH_FULL_IMAGE:figures/full_fig_p021_20.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 43 canonical work pages

  1. [1]

    https : / / github

    Flux. https : / / github . com / black - forest - labs/flux. 2, 3, 4, 6, 11

    work page

  2. [2]

    https://openai.com/chatgpt

    chatgpt. https://openai.com/chatgpt. 3, 11

    work page

  3. [3]

    Artflow: Unbiased image style transfer via re- versible neural flows

    Jie An, Siyu Huang, Yibing Song, Dejing Dou, Wei Liu, and Jiebo Luo. Artflow: Unbiased image style transfer via re- versible neural flows. In CVPR, 2021. 4, 6, 7, 14

    work page 2021

  4. [4]

    In- structpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. In- structpix2pix: Learning to follow image editing instructions. arXiv:2211.09800, 2022. 3, 6

    arXiv 2022

  5. [5]

    Internvl: Scaling up vision founda- tion models and aligning for generic visual-linguistic tasks

    Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, Bin Li, Ping Luo, Tong Lu, Yu Qiao, and Jifeng Dai. Internvl: Scaling up vision founda- tion models and aligning for generic visual-linguistic tasks. arXiv:2312.14238, 2023. 2, 5, 13

    arXiv 2023

  6. [6]

    Style injec- tion in diffusion: A training-free approach for adapting large- scale diffusion models for style transfer

    Jiwoo Chung, Sangeek Hyun, and Jae-Pil Heo. Style injec- tion in diffusion: A training-free approach for adapting large- scale diffusion models for style transfer. In CVPR, pages 8795–8805, 2024. 3, 4, 6, 7

    work page 2024

  7. [7]

    Arbitrary style transfer via multi-adaptation network

    Yingying Deng, Fan Tang, Weiming Dong, Wen Sun, Feiyue Huang, and Changsheng Xu. Arbitrary style transfer via multi-adaptation network. In ACM MM, pages 2719–2727,

    work page

  8. [8]

    Implicit style-content separation using b-lora

    Yarden Frenkel, Yael Vinker, Ariel Shamir, and Daniel Cohen-Or. Implicit style-content separation using b-lora. arXiv:2403.14572, 2024. 3, 4

    arXiv 2024

Show all 63 references
  1. [9]

    Im- age style transfer using convolutional neural networks

    Leon A Gatys, Alexander S Ecker, and Matthias Bethge. Im- age style transfer using convolutional neural networks. In ECCV, pages 2414–2423, 2016. 2, 3

    2016

  2. [10]

    Controlling perceptual fac- tors in neural style transfer

    Leon A Gatys, Alexander S Ecker, Matthias Bethge, Aaron Hertzmann, and Eli Shechtman. Controlling perceptual fac- tors in neural style transfer. In CVPR, pages 3985–3993,

  3. [11]

    Aespa-net: Aesthetic pattern-aware style transfer networks

    Kibeom Hong, Seogkyu Jeon, Junsoo Lee, Namhyuk Ahn, Kunhee Kim, Pilhyeon Lee, Daesik Kim, Youngjung Uh, and Hyeran Byun. Aespa-net: Aesthetic pattern-aware style transfer networks. In ICCV, pages 22758–22767, 2023. 4, 6, 7, 14

    2023

  4. [12]

    Diff- styler: Controllable dual diffusion for text-driven image styl- ization

    Nisha Huang, Yuxin Zhang, Fan Tang, Chongyang Ma, Haibin Huang, Weiming Dong, and Changsheng Xu. Diff- styler: Controllable dual diffusion for text-driven image styl- ization. TNNLS, 2024. 6

    2024

  5. [13]

    Arbitrary style transfer in real-time with adaptive instance normalization

    Xun Huang and Serge Belongie. Arbitrary style transfer in real-time with adaptive instance normalization. In ICCV, pages 1501–1510, 2017. 2, 3

    2017

  6. [14]

    Hq-edit: A high-quality dataset for instruction-based image editing

    Mude Hui, Siwei Yang, Bingchen Zhao, Yichun Shi, Heng Wang, Peng Wang, Yuyin Zhou, and Cihang Xie. Hq-edit: A high-quality dataset for instruction-based image editing. arXiv:2404.09990, 2024. 3, 6

    2024 arXiv

  7. [15]

    Training- free style transfer emerges from h-space in diffusion models

    Jaeseok Jeong, Mingi Kwon, and Youngjung Uh. Training- free style transfer emerges from h-space in diffusion models. arXiv:2303.15403, 3, 2023. 3

    2023 arXiv

  8. [16]

    Perceptual losses for real-time style transfer and super-resolution

    Justin Johnson, Alexandre Alahi, and Li Fei-Fei. Perceptual losses for real-time style transfer and super-resolution. In ECCV, pages 694–711. Springer, 2016. 2

    2016

  9. [17]

    Styleshot: A snap- shot on any style

    Gao Junyao, Liu Yanchen, Sun Yanan, Tang Yinhao, Zeng Yanhong, Chen Kai, and Zhao Cairong. Styleshot: A snap- shot on any style. arxiv:2407.01414, 2024. 2, 3, 4, 5, 6, 7, 14

    2024 arXiv

  10. [18]

    Style transfer by relaxed optimal transport and self-similarity

    Nicholas Kolkin, Jason Salavon, and Gregory Shakhnarovich. Style transfer by relaxed optimal transport and self-similarity. In CVPR, pages 10051–10060, 2019. 2, 3

    2019

  11. [19]

    Style- tokenizer: Defining image style by a single instance for con- trolling diffusion models

    Wen Li, Muyuan Fang, Cheng Zou, Biao Gong, Ruobing Zheng, Meng Wang, Jingdong Chen, and Ming Yang. Style- tokenizer: Defining image style by a single instance for con- trolling diffusion models. arXiv:2409.02543, 2024. 2, 3, 4, 5, 11, 12

    2024 arXiv

  12. [20]

    Universal style transfer via feature transforms

    Yijun Li, Chen Fang, Jimei Yang, Zhaowen Wang, Xin Lu, and Ming-Hsuan Yang. Universal style transfer via feature transforms. In NeurIPS, 2017. 2, 3

    2017

  13. [21]

    Visual attribute transfer through deep image analogy

    Jing Liao, Yuan Yao, Lu Yuan, Gang Hua, and Sing Bing Kang. Visual attribute transfer through deep image analogy. arXiv:1705.01088, 2017. 3

    2017 arXiv

  14. [22]

    The artbench dataset: Benchmarking generative models with art- works

    Peiyuan Liao, Xiuyu Li, Xihui Liu, and Kurt Keutzer. The artbench dataset: Benchmarking generative models with art- works. arXiv:2206.11404, 2022. 3

    2022 arXiv

  15. [23]

    Deep photo style transfer

    Fujun Luan, Sylvain Paris, Eli Shechtman, and Kavita Bala. Deep photo style transfer. InCVPR, pages 4990–4998, 2017. 2

    2017

  16. [24]

    Ava: A large-scale database for aesthetic visual analysis

    Naila Murray, Luca Marchesotti, and Florent Perronnin. Ava: A large-scale database for aesthetic visual analysis. In CVPR, pages 2408–2415. IEEE, 2012. 5, 13

    2012

  17. [25]

    Maxime Oquab, Timoth ´ee Darcet, Theo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Rus- sell Howes, Po-Yao Huang, Hu Xu, Vasu Sharma, Shang- Wen Li, Wojciech Galuba, Mike Rabbat, Mido Assran, Ni...

    2023 arXiv

  18. [26]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. In ICCV, pages 4195–4205, 2023. 2

    2023

  19. [27]

    Dead- iff: An efficient stylization diffusion model with disentan- gled representations

    Tianhao Qi, Shancheng Fang, Yanze Wu, Hongtao Xie, Ji- awei Liu, Lang Chen, Qian He, and Yongdong Zhang. Dead- iff: An efficient stylization diffusion model with disentan- gled representations. In CVPR, pages 8693–8702, 2024. 3

    2024

  20. [28]

    Learning transferable visual models from natural language supervi- sion

    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 supervi- sion. In International conference on machine learning, ...

    2021

  21. [29]

    Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation

    Nataniel Ruiz, Yuanzhen Li, Varun Jampani, Yael Pritch, Michael Rubinstein, and Kfir Aberman. Dreambooth: Fine tuning text-to-image diffusion models for subject-driven generation. In CVPR, pages 22500–22510, 2023. 2

    2023

  22. [30]

    Large-scale classifica- tion of fine-art paintings: Learning the right metric on the right feature

    Babak Saleh and Ahmed Elgammal. Large-scale classifica- tion of fine-art paintings: Learning the right metric on the right feature. arXiv:1505.00855, 2015. 3

    2015 arXiv

  23. [31]

    Viper: Visual personalization of generative models via individual preference learning

    Sogand Salehi, Mahdi Shafiei, Teresa Yeo, Roman Bach- mann, and Amir Zamir. Viper: Visual personalization of generative models via individual preference learning. arXiv:2407.17365, 2024. 5

    2024 arXiv

  24. [32]

    Styledrop: Text-to-image generation in any style

    Kihyuk Sohn, Nataniel Ruiz, Kimin Lee, Daniel Castro Chin, Irina Blok, Huiwen Chang, Jarred Barber, Lu Jiang, Glenn Entis, Yuanzhen Li, et al. Styledrop: Text-to-image generation in any style. arXiv:2306.00983, 2023. 2, 3

    2023 arXiv

  25. [33]

    Efficient example-based painting and synthesis of 2d directional texture

    Bin Wang, Wenping Wang, Huaiping Yang, and Jiaguang Sun. Efficient example-based painting and synthesis of 2d directional texture. IEEE TVCG, 10(3):266–277, 2004. 3

    2004

  26. [34]

    Instantstyle: Free lunch towards style- preserving in text-to-image generation

    Haofan Wang, Qixun Wang, Xu Bai, Zekui Qin, and Anthony Chen. Instantstyle: Free lunch towards style- preserving in text-to-image generation. arXiv:2404.02733,

    arXiv

  27. [35]

    Instantstyle-plus: Style trans- fer with content-preserving in text-to-image generation

    Haofan Wang, Peng Xing, Renyuan Huang, Hao Ai, Qixun Wang, and Xu Bai. Instantstyle-plus: Style trans- fer with content-preserving in text-to-image generation. arXiv:2407.00788, 2024. 2, 3

    2024 arXiv

  28. [36]

    Imagen editor and editbench: Advancing and evaluating text-guided image inpainting

    Su Wang, Chitwan Saharia, Ceslee Montgomery, Jordi Pont- Tuset, Shai Noy, Stefano Pellegrini, Yasumasa Onoe, Sarah Laszlo, David J Fleet, Radu Soricut, et al. Imagen editor and editbench: Advancing and evaluating text-guided image inpainting. In CVPR, pages 18359–18369, 2023. 3

    2023

  29. [37]

    Stylediffusion: Controllable disentangled style transfer via diffusion models

    Zhizhong Wang, Lei Zhao, and Wei Xing. Stylediffusion: Controllable disentangled style transfer via diffusion models. In ICCV, pages 7677–7689, 2023. 3

    2023

  30. [38]

    Omnigen: Unified image generation

    Shitao Xiao, Yueze Wang, Junjie Zhou, Huaying Yuan, Xingrun Xing, Ruiran Yan, Shuting Wang, Tiejun Huang, and Zheng Liu. Omnigen: Unified image generation. arXiv:2409.11340, 2024. 2, 6

    2024 arXiv

  31. [39]

    Csgo: Content-style composition in text-to-image genera- tion

    Peng Xing, Haofan Wang, Yanpeng Sun, Qixun Wang, Xu Bai, Hao Ai, Renyuan Huang, and Zechao Li. Csgo: Content-style composition in text-to-image genera- tion. arXiv 2408.16766, 2024. 2, 3, 4, 5, 6, 7, 14

    2024 arXiv

  32. [40]

    Freetuner: Any subject in any style with training-free diffu- sion

    Youcan Xu, Zhen Wang, Jun Xiao, Wei Liu, and Long Chen. Freetuner: Any subject in any style with training-free diffu- sion. arXiv:2405.14201, 2024. 3

    2024 arXiv

  33. [41]

    Ran Yi, Haoyuan Tian, Zhihao Gu, Yu-Kun Lai, and Paul L. Rosin. Towards artistic image aesthetics assessment: A large-scale dataset and a new method. In CVPR, pages 22388–22397, 2023. 5, 13

    2023

  34. [42]

    Magicbrush: A manually annotated dataset for instruction- guided image editing

    Kai Zhang, Lingbo Mo, Wenhu Chen, Huan Sun, and Yu Su. Magicbrush: A manually annotated dataset for instruction- guided image editing. In NeurIPS, 2023. 3

    2023

  35. [43]

    Style transfer via image component analysis

    Wei Zhang, Chen Cao, Shifeng Chen, Jianzhuang Liu, and Xiaoou Tang. Style transfer via image component analysis. IEEE TMM, 15(7):1594–1601, 2013. 3

    2013

  36. [44]

    Exact feature distribution matching for arbitrary style transfer and domain generalization

    Yabin Zhang, Minghan Li, Ruihuang Li, Kui Jia, and Lei Zhang. Exact feature distribution matching for arbitrary style transfer and domain generalization. In CVPR, pages 8035– 8045, 2022. 3

    2022

  37. [45]

    Do- main enhanced arbitrary image style transfer via contrastive learning

    Yuxin Zhang, Fan Tang, Weiming Dong, Haibin Huang, Chongyang Ma, Tong-Yee Lee, and Changsheng Xu. Do- main enhanced arbitrary image style transfer via contrastive learning. In SIGGRAPH, 2022. 4, 6, 7, 14

    2022

  38. [46]

    Prospect: Expanded condi- tioning for the personalization of attribute-aware image gen- eration

    Yuxin Zhang, Weiming Dong, Fan Tang, Nisha Huang, Haibin Huang, Chongyang Ma, Tong-Yee Lee, Oliver Deussen, and Changsheng Xu. Prospect: Expanded condi- tioning for the personalization of attribute-aware image gen- eration. arXiv:2305.16225, 2023. 2, 3

    2023 arXiv

  39. [47]

    Inversion-based style transfer with diffusion models

    Yuxin Zhang, Nisha Huang, Fan Tang, Haibin Huang, Chongyang Ma, Weiming Dong, and Changsheng Xu. Inversion-based style transfer with diffusion models. In CVPR, pages 10146–10156, 2023. 2, 3

    2023

  40. [48]

    Artbank: Artistic style transfer with pre-trained diffusion model and implicit style prompt bank

    Zhanjie Zhang, Quanwei Zhang, Wei Xing, Guangyuan Li, Lei Zhao, Jiakai Sun, Zehua Lan, Junsheng Luan, Yiling Huang, and Huaizhong Lin. Artbank: Artistic style transfer with pre-trained diffusion model and implicit style prompt bank. In AAAI, pages 7396–7404, 2024. 6, 7

    2024

  41. [49]

    Ultraedit: Instruction-based fine-grained im- age editing at scale

    Haozhe Zhao, Xiaojian Ma, Liang Chen, Shuzheng Si, Ru- jie Wu, Kaikai An, Peiyu Yu, Minjia Zhang, Qing Li, and Baobao Chang. Ultraedit: Instruction-based fine-grained im- age editing at scale. arXiv:2407.05282, 2024. 3, 6

    2024 arXiv

  42. [50]

    Transfusion: Pre- dict 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: Pre- dict the next token and diffuse images with one multi-modal model. arXiv:2408.11039, 2024. 2 Supplementary Materials...

    2024 arXiv

  43. [51]

    A swan swimming gracefully on a calm lake at sunset

  44. [52]

    A leopard resting on the branch of a tree in the jungle

  45. [53]

    A close-up of a ladybug crawling on a green leaf

  46. [54]

    A family of deer grazing in a peaceful meadow

  47. [56]

    A polar bear swimming in the icy waters of the Arctic

  48. [57]

    A family of ducks swimming in a calm pond

  49. [58]

    A majestic lion standing on a rocky ledge at sunset

  50. [59]

    A hawk swooping down to catch its prey

  51. [60]

    A close-up of a spider spinning a web

  52. [61]

    Section 1.1 delves into the process of generating content images, while Section 1.2 elaborates on the specific details of the collected style images

    More Details about OmniStyle In this section, we provide more details about OmniStyle. Section 1.1 delves into the process of generating content images, while Section 1.2 elaborates on the specific details of the collected style images. More triplet examples from our dataset a...

  53. [62]

    clip-vit-large- patch14,

    More Details about OmniFilter In this section, we provide more details about the evalua- tion schemes for content preservation, style consistency and aesthetic appeal, which are the three key aspects of the Om- niFilter. Evaluation on Content Preservation. We present the de- t...

  54. [63]

    Section 3.1 focuses on quantitative evaluations, including user studies and ablation studies

    More Experiments In this section, we present additional experimental results. Section 3.1 focuses on quantitative evaluations, including user studies and ablation studies. Section 3.2 highlights more qualitative results for further analysis. 3.1. Quantitative Experiments User ...

  55. [64]

    In contrast, the outputs generated by CAST, CSGO, and StyleShot display rich details, resulting in rel- atively higher aesthetic scores

    exhibit lower aesthetic scores due to the presence of blurred or shadowed regions in their results (visible upon magnification). In contrast, the outputs generated by CAST, CSGO, and StyleShot display rich details, resulting in rel- atively higher aesthetic scores. By comprehe...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.