Elamur Research⌕
☰
home / papers / 2025 / Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model

Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model

Paper · 2025 · Lixue Gong, Xiaoxia Hou, Fanshi Li et al.

Summary

Seedream 2.0 is a text-to-image diffusion foundation model that natively accepts prompts in both Chinese and English. It targets three weaknesses of systems such as Flux, SD3.5 and Midjourney: a skew toward one quality (e.g. aesthetics over prompt fidelity), weak rendering of long or Chinese text inside images, and a shallow grasp of Chinese cultural content. The authors build a pre-training corpus from four curated data streams, supported by a taxonomy, multimodal retrieval and active learning, plus a captioning pipeline that trades off precision and detail. A diffusion transformer is conditioned on three components: an in-house decoder-only bilingual LLM text encoder, a Glyph-Aligned ByT5 for character-level glyphs, and Scaled ROPE for unseen resolutions. Post-training combines continued training, SFT, iterative RLHF, a prompt-rewriting LLM and a refiner. Human and automatic evaluations rank it at or near the top for alignment, aesthetics, structure and text rendering, with the clearest lead on Chinese-specific prompts. The work shows that a native-language LLM encoder can carry cultural knowledge into image generation. The model already runs in Doubao and Dreamina.

Key points

  • Developed by ByteDance's Seed Vision Team (arXiv:2503.07703). It has been deployed in Doubao (豆包) and Dreamina (即梦) since early December 2024.
  • The pre-training data system has four parts: High-Quality Data, Distribution Maintenance Data (source downsampling and hierarchical cluster sampling), Knowledge Injection Data (taxonomy, multimodal retrieval and manually gathered Chinese-context pairs), and Targeted Supplementary Data.
  • Text rendering relies on Glyph-Aligned ByT5, which works especially well for long passages with complex Chinese characters. Inference is accelerated through CFG/step distillation and quantization.
  • A Chinese-characteristics benchmark rates response and aesthetics on a 1–5 scale, comparing the model against Kolors 1.5, GPT-4o, Hunyuan (Dec 2024) and MiracleVision 5.0. The largest margins appear in food, festivals, craftsmanship and architecture.
  • The model can be adapted into the instruction-based editor SeedEdit. Its human-preference alignment is reported through ELO scores, but the supplied excerpt does not contain the exact figures.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksEnglish · 2,263 worksVision & language · 657 works

Cite

Cite as Lixue Gong, Xiaoxia Hou, Fanshi Li, Liang Li, Xiaochen Lian, Fei Liu, Liyang Liu, Wei Liu, Wei Lu, Yichun Shi, Shiqi Sun, Yu Tian, Zhi Tian, Peng Wang, Xun Wang, Ye Wang, Guofeng Wu, Jie Wu, Xin Xia, Xuefeng Xiao, Linjie Yang, Zhonghua Zhai, Xinyu Zhang, Qi Zhang, Yuwei Zhang, Shijia Zhao, Jianchao Yang and Weilin Huang. 2025. Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model. arXiv:2503.07703.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2503.07703 · arXiv 2503.07703

@misc{gong2025seedream,
  title = {{Seedream 2.0: A Native Chinese-English Bilingual Image Generation Foundation Model}},
  author = {Lixue Gong, Xiaoxia Hou, Fanshi Li, Liang Li, Xiaochen Lian, Fei Liu, Liyang Liu, Wei Liu, Wei Lu, Yichun Shi, Shiqi Sun, Yu Tian, Zhi Tian, Peng Wang, Xun Wang, Ye Wang, Guofeng Wu, Jie Wu, Xin Xia, Xuefeng Xiao, Linjie Yang, Zhonghua Zhai, Xinyu Zhang, Qi Zhang, Yuwei Zhang, Shijia Zhao, Jianchao Yang and Weilin Huang},
  year = {2025},
  doi = {10.48550/arxiv.2503.07703},
  eprint = {2503.07703},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2503.07703}
}

References

44 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforcement learning. arXiv preprint arXiv:2305.13301, 2023. — [arXiv:2305.13301](https://arxiv.org/abs/2305.13301) find
  2. 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. find
  3. Mingdeng Cao, Xintao Wang, Zhongang Qi, Ying Shan, Xiaohu Qie, and Yinqiang Zheng. Masactrl: Tuning-free mutual self-attention control for consistent image synthesis and editing. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 22560–22570, October 2023. find
  4. Jingye Chen, Yupan Huang, Tengchao Lv, Lei Cui, Qifeng Chen, and Furu Wei. Textdiffuser-2: Unleashing the power of language models for text rendering. In European Conference on Computer Vision, pages 386–402. Springer, 2024. find
  5. Zhongzhi Chen, Guang Liu, Bo-Wen Zhang, Fulong Ye, Qinghong Yang, and Ledell Wu. Altclip: Altering the language encoder in clip for extended language capabilities. arXiv preprint arXiv:2211.06679, 2022. find
  6. Arpad Emmerich Elo. The proposed uscf rating system, its development, theory, and applications. Chess Life, XXII(8):242–247, 1967. find
  7. Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. In Forty-first International Conference on Machine Learning, 2024. find
  8. Shuhao Han, Haotian Fan, Jiachen Fu, Liang Li, Tao Li, Junhui Cui, Yunqiu Wang, Yang Tai, Jingwei Sun, Chunle Guo, and Chongyi Li. Evalmuse-40k: A reliable and fine-grained benchmark with comprehensive human annotations for text-to-image generation model evaluation, 2024. URL https://arxiv.org/abs/2… — [arXiv:2412.18150](https://arxiv.org/abs/2412.18150) ↗
  9. Jack Hessel, Ari Holtzman, Maxwell Forbes, Ronan Le Bras, and Yejin Choi. Clipscore: A reference-free evaluation metric for image captioning, 2022. URL https://arxiv.org/abs/2104.08718. ↗
  10. Ideogram. Ideogram. https://about.ideogram.ai/2.0, 2024. ↗
  11. Sadeep Jayasumana, Srikumar Ramalingam, Andreas Veit, Daniel Glasner, Ayan Chakrabarti, and Sanjiv Kumar. Rethinking fid: Towards a better evaluation metric for image generation, 2024. URL https://arxiv.org/abs/ 2401.09603. ↗
  12. Minchul Kim, Anil K Jain, and Xiaoming Liu. Adaface: Quality adaptive margin for face recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 18750–18759, 2022. find
  13. Black Forest Labs. Flux. https://github.com/black-forest-labs/flux, 2023. ↗
  14. Ming Li, Taojiannan Yang, Huafeng Kuang, Jie Wu, Zhaoning Wang, Xuefeng Xiao, and Chen Chen. Controlnet++: Improving conditional controls with efficient consistency feedback. In European Conference on Computer Vision, pages 129–147. Springer, 2025. find
  15. 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.
  16. Rensis Likert. A technique for the measurement of attitudes. Archives of Psychology. 140: 1–55, 1932. find
  17. 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. arXiv preprint arXiv:2404.01291, 2024. find
  18. Zeyu Liu, Weicong Liang, Zhanhao Liang, Chong Luo, Ji Li, Gao Huang, and Yuhui Yuan. Glyph-byt5: A customized text encoder for accurate visual text rendering. In European Conference on Computer Vision, pages 361–377. Springer, 2024. find
  19. Zeyu Liu, Weicong Liang, Yiming Zhao, Bohan Chen, Lin Liang, Lijuan Wang, Ji Li, and Yuhui Yuan. Glyph-byt5-v2: A strong aesthetic baseline for accurate multilingual visual text rendering. arXiv preprint arXiv:2406.10208, 2024. — [arXiv:2406.10208](https://arxiv.org/abs/2406.10208) find
  20. Meitu. Meitu. https://www.whee.com/ai/text-to-image, 2024. ↗
  21. Midjourney. Midjourney v6.1. https://www.midjourney.com/, 2024. ↗
  22. OpenAI, :, Aaron Hurst, and Adam Lerer et al. Gpt-4o system card, 2024. URL https://arxiv.org/abs/2410. 21276. ↗
  23. Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023.
  24. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8763. PMLR, 20…
  25. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020.
  26. Recraft. Recraft v3. https://www.recraft.ai/projects, 2024. ↗
  27. Yuxi Ren, Xin Xia, Yanzuo Lu, Jiacheng Zhang, Jie Wu, Pan Xie, Xing Wang, and Xuefeng Xiao. Hyper-sd: Trajectory segmented consistency model for efficient image synthesis. Advances in Neural Information Processing Systems, 37:117340–117362, 2025. find
  28. Yichun Shi, Peng Wang, and Weilin Huang. Seededit: Align image re-generation to image editing, 2024. URL https://arxiv.org/abs/2411.06686. — [arXiv:2411.06686](https://arxiv.org/abs/2411.06686) ↗
  29. Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. Roformer: Enhanced transformer with rotary position embedding. Neurocomputing, 568:127063, 2024.
  30. Kolors Team. Kolors: Effective training of diffusion model for photorealistic text-to-image synthesis. arXiv preprint, 2024. find
  31. Tencent. Hunyuan. https://console.cloud.tencent.com/hunyuan/experience/image, 2024. ↗
  32. Yuxiang Tuo, Wangmeng Xiang, Jun-Yan He, Yifeng Geng, and Xuansong Xie. Anytext: Multilingual visual text generation and editing. arXiv preprint arXiv:2311.03054, 2023. — [arXiv:2311.03054](https://arxiv.org/abs/2311.03054) find
  33. Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik. Diffusion model alignment using direct preference optimization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, p… find
  34. Shaojin Wu, Fei Ding, Mengqi Huang, Wei Liu, and Qian He. Vmix: Improving text-to-image diffusion model with cross-attention mixing control. arXiv preprint arXiv:2412.20800, 2024. — [arXiv:2412.20800](https://arxiv.org/abs/2412.20800) find
  35. 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. arXiv preprint arXiv:2306.09341, 2023. — [arXiv:2306.09341](https://arxiv.org/abs/2306.09341) find
  36. Jiazheng Xu, Xiao Liu, Yuchen Wu, Yuxuan Tong, Qinkai Li, Ming Ding, Jie Tang, and Yuxiao Dong. Imagereward: Learning and evaluating human preferences for text-to-image generation. Advances in Neural Information Processing Systems, 36, 2024. find
  37. Linting Xue, Aditya Barua, Noah Constant, Rami Al-Rfou, Sharan Narang, Mihir Kale, Adam Roberts, and Colin Raffel. Byt5: Towards a token-free future with pre-trained byte-to-byte models. Transactions of the Association for Computational Linguistics, 10:291–306, 2022. find
  38. Hu Ye, Jun Zhang, Sibo Liu, Xiao Han, and Wei Yang. Ip-adapter: Text compatible image prompt adapter for text-to-image diffusion models. arXiv preprint arXiv:2308.06721, 2023. — [arXiv:2308.06721](https://arxiv.org/abs/2308.06721) find
  39. Jiahui Yu, Yuanzhong Xu, Jing Yu Koh, Thang Luong, Gunjan Baid, Zirui Wang, Vijay Vasudevan, Alexander Ku, Yinfei Yang, Burcu Karagol Ayan, Ben Hutchinson, Wei Han, Zarana Parekh, Xin Li, Han Zhang, Jason Baldridge, and Yonghui Wu. Scaling autoregressive models for content-rich text-to-image generat… — [arXiv:2206.10789](https://arxiv.org/abs/2206.10789) find
  40. Yu Zeng, Vishal M Patel, Haochen Wang, Xun Huang, Ting-Chun Wang, Ming-Yu Liu, and Yogesh Balaji. Jedi: Joint-image diffusion models for finetuning-free personalized text-to-image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 6786–6795, 2024. find
  41. Jiacheng Zhang, Jie Wu, Weifeng Chen, Yatai Ji, Xuefeng Xiao, Weilin Huang, and Kai Han. Onlinevpo: Align video diffusion model with online video-centric preference optimization. arXiv preprint arXiv:2412.15159, 2024. — [arXiv:2412.15159](https://arxiv.org/abs/2412.15159) find
  42. Jiacheng Zhang, Jie Wu, Yuxi Ren, Xin Xia, Huafeng Kuang, Pan Xie, Jiashi Li, Xuefeng Xiao, Min Zheng, Lean Fu, et al. Unifl: Improve stable diffusion via unified feedback learning. arXiv preprint arXiv:2404.05595, 2024. — [arXiv:2404.05595](https://arxiv.org/abs/2404.05595) find
  43. Sixian Zhang, Bohan Wang, Junqiang Wu, Yan Li, Tingting Gao, Di Zhang, and Zhongyuan Wang. Learning multi-dimensional human preference for text-to-image generation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8018–8027, 2024. find
  44. Yanli Zhao, Andrew Gu, Rohan Varma, Liang Luo, Chien-Chin Huang, Min Xu, Less Wright, Hamid Shojanazeri, Myle Ott, Sam Shleifer, et al. Pytorch fsdp: experiences on scaling fully sharded data parallel. arXiv preprint arXiv:2304.11277, 2023. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/gong-2025-seedream-2-0-native/