HKUST Computer Science & Engineering

Hongyu Liu

PhD Candidate

hongyu/hongyu-forest-portrait.webp

Computer Vision · Computer Graphics · Generative AI

I am a Ph.D. student in Computer Science and Engineering at HKUST, advised by Prof. Qifeng Chen.

My research spans video generation and generative world models, human-centered multimodal spatial intelligence, and image generation and editing. I currently focus on post-training world models and embodied intelligence.

Outside the lab, I love fishing and hope to explore China’s rivers, lakes, and coastlines, one fishing trip at a time.

News

Nov 20, 2026 We released LongCat-Video-Avatar 1.6! By adapting OPSD-V for audio-driven avatar continuation, the model generates coherent avatar videos up to 10 minutes long, with more natural expressions and reduced error accumulation. Code · Technical report.
Oct 27, 2026 We released SMD, a simple framework for few-step diffusion distillation across image, video, audio–video, and 3D generation. Code · Paper.
Oct 09, 2026 Two papers have been accepted to NeurIPS 2026!
Oct 06, 2026 We released WorldSonus, a streaming video-to-audio model for synchronized, controllable spatial sound.
Sep 03, 2026 We released HelixWorld, a real-time interactive audio-visual world model with synchronized video and spatial sound.

Recent Highlights

Meituan LongCat · Technical Report

LongCat-Video-Avatar 1.6

Expressive audio-driven avatars with OPSD-V-based post-training for coherent video generation up to 10 minutes.

MeiGen · Technical Report

OPSD-V

On-policy self-distillation for post-training few-step autoregressive video generators, reducing long-horizon degradation.

NVIDIA · NeurIPS 2026

DyaPlex

Full-duplex speech-motion model for dyadic interaction, coupling streaming speech and body motion for responsive digital agents.

Highlighted Research Topics

Three connected directions: video generation and generative world models, human-centered multimodal spatial intelligence, and image generation and editing.

All publications