Excited to share that
@Theta_Network’s joint research with Prof. Xiao
@ZhenXiaoCornell’s team has been presented at CVPR 2026! 🎉
📄 Yo'City: Personalized and Boundless 3D Realistic City Scene Generation via Self-Critic Expansion
Paper:
openaccess.thecvf.com/conten…
CVPR
@CVPR is the premier conference in computer vision and AI, showcasing the latest advances in visual intelligence, generative AI, robotics, and digital world creation.
Our paper presents Yo’City, a novel agentic framework for generating personalized, realistic, and infinitely expandable 3D city environments from natural language instructions.
Unlike traditional approaches that rely on handcrafted rules, maps, or training specialized city-generation models, Yo’City leverages the reasoning and planning capabilities of large AI models to create city-scale virtual worlds through a hierarchical generation process.
Yo’City introduces:
-- Hierarchical City Planning: A "City–District–Grid" framework where a Global Planner designs the overall city structure and a Local Designer refines each district with detailed architectural and spatial layouts.
-- Agentic 3D Generation: A produce–refine–evaluate generation loop that continuously improves visual fidelity, semantic consistency, and structural realism before converting scenes into 3D assets.
-- Boundless City Expansion: A relationship-guided expansion mechanism that enables cities to grow continuously through user interaction while maintaining spatial coherence and semantic consistency.
-- Personalized World Creation: Support for highly customized environments ranging from realistic metropolitan areas to themed worlds such as fantasy parks, innovation districts, and virtual commercial centers.
Experimental results show that Yo’City consistently outperforms existing state-of-the-art approaches across semantic consistency, geometric fidelity, texture quality, layout coherence, scene coverage, and overall realism.
This work aligns closely with Theta’s long-term vision for AI-powered virtual worlds, digital twins, and immersive 3D experiences running on decentralized GPU infrastructure. As generative AI evolves from generating images and videos to generating entire interactive worlds, scalable AI compute will become a foundational layer of the next generation of digital experiences.
As always,
@Theta_Network remains committed to advancing the intersection of AI, 3D content generation, and decentralized infrastructure. Stay tuned for more breakthroughs! 🌐