Generating 360 Video is What You Need For a 3D Scene
WorldPrompter generates high-quality Gaussian splat 3D scenes by leveraging 360° video as an intermediate representation generated from text prompts.
About Me
I am Zhaoyang Zhang, a 4th year Ph.D. student at Yale University advised by Prof. Julie Dorsey. I am affiliated with the Yale Computer Graphics Group. Before joining Yale, I obtained my B.Eng. at University of Chinese Academy of Sciences, with highest honor. My undergrad research was advised by Prof. Lin Gao.
My research specializes in generative models for computer graphics, with a focus on developing geometrically consistent methods for digital content creation, spanning 3D scenes, videos, and beyond. I am fortunate to have interned at Apple in 2025 and Adobe🔺 in 2024.
✉️ Email: zhaoyang dot zhang at yale dot edu
Update Sep 2025
I am actively seeking for 2026 summer internships and research collaborations on the following topics:
- 3D Scene Generation/Reconstruction
- Long/Panoramic Video Generation
- MLLM-based Visual Generation
Feel free to reach out 🙂
Education
Ph.D. in Computer Science
Yale University
M.Phil. in Computer Science
Yale University
B.Eng. in Computer Science and Technology
University of Chinese Academy of Sciences
High School
Nanjing Foreign Language School
Interests
Computer Graphics
3D Computer Vision
Scene/Video Generation
Multimodal Large Language Models (MLLMs)
Experience
Research Intern
Apple Research Intern
Adobe🔺
Education
Ph.D. in Computer Science
Yale UniversityM.Phil. in Computer Science
Yale UniversityB.Eng. in Computer Science and Technology
University of Chinese Academy of SciencesHigh School
Nanjing Foreign Language School
Publications
Predicting Fabric Appearance Through Thread Scattering and Inversion
We introduce a differentiable pipeline that digitizes physical threads using a cost-efficient multi-view capture setup and a novel thread scattering model, enabling accurate …
Mengqi (Mandy) Xia
•
•
1 min read
All-day Depth Completion
We introduces SpaDe, a plug-and-play depth estimation method that fuses sparse LiDAR depth with camera images under varying illumination, using synthetic data to learn coarse dense …
Vadim Ezhov
•
•
1 min read
Text2Face: Text-Based Face Generation With Geometry and Appearance Control
We present Text2Face, a local-to-global framework with geometry- and appearance-aware graph neural networks that model dependencies among facial parts to resolve text-to-face …