Researcher · Generative Intelligence
JD Future Academy · Tech Genius Team (TGT)
Curiosity fuels discovery.
Persistence unlocks the unknown.
My research focuses on large-scale audio-visual generative models and interactive world models for games and real-world environments.
I am currently a Researcher at JD Future Academy, JD.com, as a member of the Tech Genius Team (TGT).
I received my Master’s degree in Computer Technology from Tsinghua University in 2026, under the supervision of Prof. Chun Yuan. I obtained my Bachelor’s degree in Computer Science and Technology from the Yingcai Honors College at the University of Electronic Science and Technology of China in 2023, where I was fortunate to be advised by Prof. Xile Zhao.
Previously, I worked as a Research Assistant at MMLab, The Chinese University of Hong Kong (CUHK), under the supervision of Prof. Tianfan Xue.
My research focuses on large-scale audio-visual generative models and interactive world models for games and real-world environments.
Recent releases, publications, and research milestones.
Research spanning persistent audio-visual worlds, long-form video generation, visual editing, and restoration.
* indicates equal contribution.
JoyAI‑Echo‑1.5 is a unified audio‑visual generation system featuring two purpose‑built variants: a long‑video variant with composable cross‑shot memory for sustained identity consistency, and a world‑model variant with geometry‑aware 6‑DoF camera control for interactive viewpoint navigation.
An omnimodal world model for generative media that responds to continuous navigation while video, environmental sound, music, and speech evolve together. I proposed Short-Horizon and Long-Horizon Audio-Visual Self-Gradient Forcing and was responsible for EchoWM’s causal training.
Self Gradient Forcing (SGF) recovers the missing context-gradient path for self-generated causal memory through a bounded two-pass replay, enabling models trained with only a 5-second window to extrapolate to minute-scale videos with stronger identity, layout, and temporal stability. It also supports the causal training of EchoWM.
FlashVSR is a streaming, one-step diffusion-based video super-resolution framework with block-sparse attention and a Tiny Conditional Decoder. It reaches ~17 FPS at 768×1408 on a single A100 GPU. A Locality-Constrained Attention design further improves generalization and perceptual quality on ultra-high-resolution videos.
Cobra is a novel efficient long-context fine-grained ID preservation framework for line art colorization, achieving high precision, efficiency, and flexible usability for comic colorization. By effectively integrating extensive contextual references, it transforms black-and-white line art into vibrant illustrations.
PowerPaint is the first versatile image inpainting model that simultaneously achieves state-of-the-art results in various inpainting tasks such as text-guided object inpainting, context-aware image inpainting, shape-guided object inpainting with controllable shape-fitting, and outpainting.
BrushEdit is an all-in-one image inpainting and editing framework that combines multimodal large language models (MLLMs) with the enhanced dual-branch diffusion inpainting model BrushNetX. It supports free-form instruction-guided interactive editing, achieves superior performance in background preservation and text alignment, and provides a user-friendly multi-round editing experience.
Safe-Sora: a framework for embedding graphical watermarks into video generation, achieving state-of-the-art quality, fidelity, and robustness through hierarchical adaptive matching and a 3D wavelet-enhanced Mamba architecture.
ColorFlow is the first model designed for fine-grained ID preservation in image sequence colorization, utilizing contextual information. Given a reference image pool, ColorFlow accurately generates colors for various elements in black and white image sequences, including the hair color and attire of characters, ensuring color consistency with the reference images.
We propose the UConNet for image and video deraining. Our UConNet learns a relationship between trade-off parameters of the loss function and weightings of feature maps. At the inference stage, the weightings can be adaptively controlled to handle different rain scenarios, resulting in high generalization abilities. Extensive experimental results validate the effectiveness, generalization abilities, and efficiency of UConNet.
Research across generative video, efficient diffusion, visual restoration, and editing.
Large-scale audio-visual generative models and interactive world models.
Supervised by Shi Guo and Tianfan Xue.
Topics: Video Super-Resolution · Diffusion Acceleration · Sparse Attention.
Supervised by Zhaoyang Zhang and Ying Shan.
Topics: Comic Colorization · Video Generation · Diffusion.
Supervised by Yanhong Zeng and Kai Chen.
Topics: Image Inpainting · Diffusion.
Advised by Prof. Chun Yuan. Previously a Research Assistant at MMLab, CUHK, advised by Prof. Tianfan Xue.
Advised by Prof. Xile Zhao.
Academic distinctions received during undergraduate and graduate study.
Real-time visitor locations and traffic to this homepage.
