Hugging Face has introduced Infinite London, an AI-generated 3D environment that creates endless, explorable cities using Wave Function Collapse (WFC). The project allows users to explore procedurally generated Victorian cities composed of 36 AI-generated 3D tiles. Each visit to the environment results in a new, coherent urban landscape, with streets, buildings, and landmarks dynamically assembled as users navigate the space using WASD controls. The system leverages modular tilesets and WFC to generate infinite variations of London while maintaining structural coherence. The result is a continuously evolving cityscape that expands as users move through it. Source: huggingface
The project utilizes a combination of AI-generated tiles and advanced reconstruction techniques to create textured 3D models. Each tile is generated using ideogram-ai/ideogram4, which produces isometric game-kit specimens on black backgrounds. These images are then reconstructed into textured GLB files using microsoft/TRELLIS.2, followed by simplification for efficient browser rendering. The entire process is driven by an agent that follows the specifications in agents.md, allowing users to replicate the workflow by copying the instructions directly into their coding agents. The system caches the 36 expensive tile generations, enabling the creation of infinite cities at minimal cost. Source: huggingface
The project's success hinges on the design of the 36 tiles, which follow a specific socket grammar. Each tile's edges are either open paving or street, and WFC ensures that only tiles with matching edges are placed together. The tileset includes 10 ground types, 12 street configurations, 9 building styles, 3 rare landmarks, and 2 street end-caps. Every tile is an inspectable GLB file on the Hugging Face Hub, allowing users to examine or reuse individual components. The system also features a browser-based viewer using three.js, with chunked infinite WFC that generates the world in 8×8-cell segments around the camera. Source: huggingface
