Zilyu Ye (叶子绿)

Hi, I’m Zilyu Ye (叶子绿), a researcher at ByteDance Seed. My research focuses on multimodal learning and post-training for visual generation, with a particular interest in reinforcement learning, reward modeling, and diffusion models.

I studied Artificial Intelligence at the School of Future Technology, South China University of Technology, from 2022 to 2026. Before joining ByteDance Seed as a researcher, I interned at Seed from 2025 to 2026 and at the School of Engineering, Westlake University, from 2024 to 2025. During my time at Westlake, I had the privilege of working with Prof. Qi Liu and Prof. Guo-Jun Qi.

news

Mar 17, 2025	Started internship at Bytedance Seed Vision Team.
Mar 14, 2024	Started internship at MAPLE Lab, Westlake University.

selected publications

CVPR
Schedule On the Fly: Diffusion Time Prediction for Faster and Better Image Generation

Zilyu Ye, Zhiyang Chen^†, Tiancheng Li, Zemin Huang, Weijian Luo, and Guo-Jun Qi^†

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025

Abs Bib PDF Code

Diffusion and flow matching models have achieved remarkable success in text-to-image generation. However, these models typically rely on the predetermined denoising schedules for all prompts. The multi-step reverse diffusion process can be regarded as a kind of chain-of-thought for generating high-quality images step by step. Therefore, diffusion models should reason for each instance to adaptively determine the optimal noise schedule, achieving high generation quality with sampling efficiency. In this paper, we introduce the Time Prediction Diffusion Model (TPDM) for this. TPDM employs a plug-and-play Time Prediction Module (TPM) that predicts the next noise level based on current latent features at each denoising step. We train the TPM using reinforcement learning to maximize a reward that encourages high final image quality while penalizing excessive denoising steps. With such an adaptive scheduler, TPDM not only generates high-quality images that are aligned closely with human preferences but also adjusts diffusion time and the number of denoising steps on the fly, enhancing both performance and efficiency. With Stable Diffusion 3 Medium architecture, TPDM achieves an aesthetic score of 5.44 and a human preference score (HPS) of 29.59, while using around 50% fewer denoising steps to achieve better performance.
@inproceedings{ye2025scheduleflydiffusiontime, title = {Schedule On the Fly: Diffusion Time Prediction for Faster and Better Image Generation}, author = {Ye, Zilyu and Chen, Zhiyang and Li, Tiancheng and Huang, Zemin and Luo, Weijian and Qi, Guo-Jun}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, year = {2025}, }

arXiv

Seedance 1.0: Exploring the Boundaries of Video Generation Models

Yu Gao, Haoyuan Guo, Tuyen Hoang, Weilin Huang, Lu Jiang, Fangyuan Kong, and 38 more authors

In , 2025

Bib

@inproceedings{gao2025seedance10exploringboundaries,
  title = {Seedance 1.0: Exploring the Boundaries of Video Generation Models},
  author = {Gao, Yu and Guo, Haoyuan and Hoang, Tuyen and Huang, Weilin and Jiang, Lu and Kong, Fangyuan and Li, Huixia and Li, Jiashi and Li, Liang and Li, Xiaojie and Li, Xunsong and Li, Yifu and Lin, Shanchuan and Lin, Zhijie and Liu, Jiawei and Liu, Shu and Nie, Xiaonan and Qing, Zhiwu and Ren, Yuxi and Sun, Li and Tian, Zhi and Wang, Rui and Wang, Sen and Wei, Guoqiang and Wu, Guohong and Wu, Jie and Xia, Ruiqi and Xiao, Fei and Xiao, Xuefeng and Yan, Jiangqiao and Yang, Ceyuan and Yang, Jianchao and Yang, Runkai and Yang, Tao and Yang, Yihang and Ye, Zilyu and Zeng, Xuejiao and Zeng, Yan and Zhang, Heng and Zhao, Yang and Zheng, Xiaozheng and Zhu, Peihao and Zou, Jiaxin and Zuo, Feilong},
  year = {2025},
  eprint = {2506.09113},
  archiveprefix = {arXiv},
  primaryclass = {cs.CV},
  url = {https://arxiv.org/abs/2506.09113}
}