PhD Student · Generative AI · Video Systems

Luozhou Wang (王罗州)

I am a fourth-year PhD student in Artificial Intelligence at HKUST(GZ), advised by Prof. Ying-Cong Chen. I work on controllable and efficient generative models, with a recent focus on long-video generation infrastructure. I am also collaborating closely with Yukang Chen on efficient training and inference systems for video generation.

Luozhou Wang

News

We released LongLive 2.0, an open-source NVFP4 parallel infrastructure for long-video generation, covering AR training, few-step distillation, multi-shot training and inference, SP acceleration, NVFP4 KV cache, and asynchronous VAE decoding.

The transparent video generation technology behind TransPixeler has been transferred into Adobe Firefly, enabling transparent-background video generation with foreground and alpha video exports.

Research

My research interests center on controllable visual generation, video customization, transparent RGBA video generation, and practical systems for efficient long-video model training and deployment.

Selected Publications

LongLive 2.0 NVFP4 key result overview

New release

LongLive-2.0: An NVFP4 Parallel Infrastructure for Long Video Generation

Yukang Chen*, Luozhou Wang*, Wei Huang*, Shuai Yang*, Bohan Zhang, Yicheng Xiao, Ruihang Chu, Weian Mao, Qixin Hu, Shaoteng Liu, Yuyang Zhao, Huizi Mao, Ying-Cong Chen, Enze Xie, Xiaojuan Qi, Song Han

arXiv, 2026 · * equal contribution

A 4-bit long-video generation infrastructure built around NVFP4 and parallelism, supporting the full path from training to deployment and reaching 45.7 FPS on LongLive-2.0-5B.

Paper Code Demo