About me
I am a researcher at StepFun, working with Dr. Gang Yu on advancing AI-generated content (AIGC), personalized content creation, 3D generation and computer graphics. I am focused on applying advanced AI techniques within the field of creative content to explore its potential applications. Prior to joining StepFun, I held research roles at Tencent, SenseTime Research, and Shanghai AI Lab.
Selected Projects
-
Step1X-Edit & GEdit-Bench
Step1X-Edit is an open-source general editing model that achieves proprietary-level performance with comprehensive editing capabilities.
GEdit-Bench is benchmark that evaluate editing models with genuine user instructions.
>> Project Page -
OmniSVG
OmniSVG is a family of SVG generation models which built on a pre-trained vision-language model Qwen-VL and incorporates an SVG tokenizer. It is capable of progressively generating high-quality SVGs across a wide spectrum of complexity — from simple icons to intricate anime characters. It demonstrates remarkable versatility through multiple generation modalities, including Text-to-SVG, Image-to-SVG, and Character-Reference SVG, making it a powerful and flexible solution for diverse creative tasks.
>> Project Page -
WithAnyone
WithAnyone tackles the copy-paste problem in identity-consistent generation by introducing a large-scale paired dataset (MultiID-2M) and a contrastive training paradigm that balances fidelity and diversity. It achieves high identity preservation while enabling controllable and expressive image generation.
>> Project Page -
OmniLottie
OmniLottie generates high-quality vector animations from multi-modal instructions via parameterized Lottie tokens, and is trained on MMLottie-2M.
>> Project Page -
MVPaint
MVPaint explores synchronized multi-view diffusion to create consistent and detailed 3D textures from textual descriptions. With synchronized multi-view diffusion, it delivers seamless, high-res textures with minimal UV wrapping dependency.
>> Project Page -
MeshXL
MeshXL is a family of generative pre-trained foundation models for 3D mesh generation. With the Neural Coordinate Field representation, the generation of unstructured 3D mesh data can be seaminglessly addressed by modern LLM methods.
>> Project Page -
DNA-Rendering
DNA-Rendering is a large-scale, high-fidelity repository of human performance data for neural actor rendering, which contains large volume of data with diverse attributes and rich annotation. Along with the dataset, a large-scale and quantitative benchmark in full-scale with multiple tasks on human rendering is provided.
>> Project Page -
MonoHuman
MonoHuman, which robustly renders view-consistent and high-fidelity avatars under arbitrary novel poses from monocular videos. The key insight is to model the deformation field with bi-directional constraints and explicitly leverage the off-the-peg keyframe information to reason the feature correlations for coherent results. Extensive experiments demonstrate the superiority of our proposed MonoHuman over state-of-the-art methods.
>> Project Page
