Shenghai Yuan

Second Year Master @Peking University

👋 About Me

Greetings!

I am currently a second-year master of computer science and technology at School of Electron and Computer Engineering, Peking University, advised by Prof. Li Yuan. I received my B. Eng in School of Computer Science and Technology, Guangdong University of Technology. My research interests include but are not limited to Video Generation, Generative AI, and Efficient AI.

🔥 News

Jul 2025 🎉 Joined ByteDance as a Research Intern.
Dec 2024 🎉 I received my B.E. degree from Guangdong University of Technology (GDUT), awarded the Outstanding Graduate, ranking 1^st/815 in overall scores for four years (2020-2024)!

🎨 Selected Works

🌟 Helios: Real Real-Time Long Video Generation Model

Shenghai Yuan, et al. (2026)

The first 14B video generation model that achieves minute-scale, high-quality video synthesis at 19.5 FPS on a single H100 GPU (about 10 FPS on a single Ascend NPU) — without relying on conventional long video anti-drifting strategies or standard video acceleration techniques.

[Paper] [Code] [Page] [HuggingFace]

🌟 Identity-Preserving Text-to-Video Generation by Frequency Decomposition

Shenghai Yuan, et al. (CVPR 2025 Highlight🌟)

A tuning-free DiT-based controllable IPT2V model to keep human-identity consistent in the generated video. The approach draws inspiration from previous studies on frequency analysis.

[Paper] [Code] [Page] [Poster] [HuggingFace]

🌟 Open-Sora-Plan

Open-Sora-Plan Team, Shenghai Yuan (2024)

Pioneering open-source effort to reproduce Sora. Responsible for the high-resolution, long-duration video generation architecture and curating open-source high-quality video datasets.

[Paper] [Code] [HuggingFace]

🌟 MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators

Shenghai Yuan, et al. (TPAMI 2025)

A metamorphic video generation model based on the given prompts. The main idea is to enhance the capacity of generation models to accurately depict the real world through our proposed methods.

[Paper] [Code] [Page] [Poster] [HuggingFace]