Selected Publications

OpenS2V-Nexus: A Detailed Benchmark and Million-Scale Dataset for Subject-to-Video Generation

Shenghai Yuan^*, Xianyi He^*, Yufan Deng, YangYe, Jinfa Huang, Bin Lin, Jiebo Luo, Li Yuan

Neural Information Processing Systems (NeurIPS D&B), 2025

Paper Code Page Dataset Benchmark Leaderboard Poster

Introducing OpenS2V-Nexus which consists of: (i) OpenS2V-Eval, a fine-grained benchmark, and (ii) OpenS2V-5M, a million-scale dataset.

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

Shenghai Yuan, Jinfa Huang, Xianyi He, Yunyang Ge, Yujun Shi, Liuhan Chen, Jiebo Luo, Li Yuan

IEEE Conference on Computer Vision and Pattern Recognition (CVPR Highlight), 2025

Paper Code Page Dataset Video Poster

We present ConsisID, an identity-preserving text-to-video generation model, which can keep human-identity consistent in the generated video.

ChronoMagic-Bench : A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation

Shenghai Yuan, Jinfa Huang, Yongqi Xu, YaoYang Liu, Shaofeng Zhang, Yujun Shi, Ruijie Zhu, Xinhua Cheng, Jiebo Luo, Li Yuan

Neural Information Processing Systems (NeurIPS D&B Spotlight), 2024

Paper Code Page LeaderBoard Dataset_Pro Dataset_ProH Poster

We present ChronoMagic-Bench, a benchmark for metamorphic evaluation of text-to-time-lapse video generation, can reflect the physical prior capacity of the T2V model.

MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators

Shenghai Yuan^*, Jinfa Huang^*, Yujun Shi, Yongqi Xu, Ruijie Zhu, Bin Lin, Xinhua Cheng, Li Yuan, Jiebo Luo

IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2025

Paper Code Page Dataset Poster

We are thrilled to present MagicTime, a metamorphic time-lapse video generation model and a new dataset ChronoMagic, support U-Net or DiT-based T2V frameworks.

LHNetV2: A Balanced Low-cost Hybrid Network for Single Image Dehazing

Shenghai Yuan, Jijia Chen, Wenchao Jiang, Zhiming Zhao, Song Guo

IEEE Transactions on Multimedia (TMM), 2024

Paper Code

We introduce LHNetV2, an advanced version of LHNet that combines various dehazing features more efficiently while enhancing inference speed.

LHNet: A Low-cost Hybrid Network for Single Image Dehazing

Shenghai Yuan, Jijia Chen, Jiaqi Li, Wenchao Jiang, Song Guo

ACM International Conference on Multimedia (ACM MM), 2023

Paper Code Poster

We propose LHNet, a Low-cost Hybrid Network that effectively merges various features for single image dehazing.

Selected Projects

Open-Sora Plan: Open-source large video generation model

PKU-Yuan Lab and Tuzhan AI etc.,

Report Code

SparkWan: Few-Step is Enough for Video Diffusion Transformer via Hybrid Adversarial Post-Training

Zongjian Li*, Shenghai Yuan*, Li Yuan

Report Code

Patents

An acceleration system and method for deconvolution calculation in neural networks

Shenghai Yuan, Bosheng Liu, Xiaoming Lie, Guanfu Cai

CN202210582998.5 / CN114821262A

A cleaning device for exterior windows of high-rise buildings

Shenghai Yuan

CN201821848303.9 / CN209678371U

FPGA-based mixed-precision data frequency domain convolution acceleration method and system

Yongqi Xu, Bosheng Liu, Yi Chen, Shenghai Yuan

CN201821848303.9 / CN209678371U

Awards

National Scholarship Award, PRC (2025)
Pacemaker to Merit Student, Peking University (2025)
National Scholarship Award, PRC (2023)
National Scholarship Award, PRC (2022)
National Scholarship Award, PRC (2021)

Academic Service

ICCV 2025, NeurIPS 2025, CVPR 2025, SIGGRAPH 2025, ACM MM 2024, ACM MM 2023