把视频生成做成一个可训练、可评估、可迭代的系统。 Building video generation as a system that can learn, verify, and improve.
Video foundation model post-training, reward modeling, reinforcement learning
我目前是 字节跳动 Seed 的资深研究员，主要关注多模态生成，尤其是 R2V 与 Post-Training。代表工作包括 Seedance 2.5 和 Seedance 2.0。
I am currently a Senior Researcher at ByteDance Seed, focusing on multimodal generation, especially R2V and post-training. Representative work includes Seedance 2.5 and Seedance 2.0.
此前，我在字节跳动智能创作实验室从事视频生成与智能创作研究，参与了 Phantom（1.5k★）、HuMo（1.3k★） 和 DreaMontage 等工作。
Previously, I worked in ByteDance's Intelligent Creation Lab on video generation and intelligent creation. I contributed to projects including Phantom (1.5k★), HuMo (1.3k★), and DreaMontage.
我本科毕业于华中科技大学，硕士毕业于中国科学院自动化研究所。2017 年硕士毕业后加入字节跳动 AI Lab，从事视频理解研究；此后参与了一键成片等智能创作能力建设，并支持 剪映 等产品功能。
I received my B.Eng. from HUST and my M.Eng. from CASIA. After graduating in 2017, I joined ByteDance AI Lab to work on video understanding, and later contributed to intelligent-creation features such as one-click video creation for Jianying.
参与 Seaweed / Seedance 1.0 的数据与 SFT、Seedance 1.5 pro 的 RM 与 RL、Seedance 2.0 的多模态 R2V RL 与 Reward Model，以及 Seedance 2.5 的 R2V RL 与 rubric verifier。Contributed to data and SFT for Seaweed / Seedance 1.0, reward modeling and RL for Seedance 1.5 pro, multimodal R2V RL and reward modeling for Seedance 2.0, and R2V RL with rubric verifiers for Seedance 2.5.
参与跨场景主体一致性数据构建，通过主体检测、跨库检索与 identity verification 获得高质量 cross-pair。Contributed to cross-context subject-consistency data through detection, cross-corpus retrieval, and identity verification.
Mingzhen Sun*, Weining Wang*, Gen Li, Jiawei Liu, et al.
大规模短视频理解与检索工作。Earlier work in large-scale short-video understanding and retrieval.
Gen Li, Shikun Xu, Xiang Liu, Lei Li, Changhu Wang.
我也对借助当下的 AI Agent 构建 Quant、Personal Health 和 Movie Studio 等应用感兴趣，欢迎联系。 I am also interested in using today's AI agents to build applications for Quant, Personal Health, and Movie Studio. Feel free to reach out.
