AI swarms build 3D worlds

Autonomous Claude Fable 5.1 agent swarms build explorable, browser-native reconstructions of real-world locations. The entire pipeline runs without traditional game engines or proprietary 3D tile sets.

“Every building, storefront, sign, tree and traffic light is generated from open data and public reference imagery by code”

Union Square comes to life

The flagship world recreates San Francisco's Union Square with 129 storefronts, 220 pedestrians, and cable cars. It includes day-to-night cycles and interactive interiors for Apple and Nintendo stores.

“129 identified storefronts, working traffic lights and cable cars, day/sunset/night, and two explorable interiors”

Pure Three.js powered by open data

Projects run locally as plain Three.js apps using open data sources like OpenStreetMap and USGS elevation. Anyone can clone the repo and launch a world with standard npm commands.

“No game engine. No proprietary 3D tiles.”

Automated end-to-end generation pipeline

Research agents pull spatial specs before headless Blender scripts generate optimized 3D asset kits. The browser runtime then dynamically assembles terrain, props, and crowds from JSON specs.

Visual QA powered by agent reviewers

Playwright captures screenshots from identical camera angles to diff them against real photos. Specialized AI reviewer agents analyze the diffs and file reports that drive fix cycles.

“34 camera-matched viewpoints vs. real photographs · 147 comparison sheets · 9 independent reviewer reports”