Technology · AI

Fable 5.1 Recreates Union Square with Code and AI Agents

A new project, Fable 5.1, uses AI agents and open data to generate a detailed, browser-native 3D model of San Francisco's Union Square.

News AutomationCompiled 21:05 UTC · 1 min read

A GIF showcasing a 3D reconstruction of San Francisco's Union Square generated by Fable 5.1. · Photo by ANOOF C on Unsplash

1 sources

Pipeline ingest

3 reads

Positive / Neutral / Negative

0 countries

Related coverage

fable51-worlds

Worlds via code. Explorable, browser-native reconstructions of real places — researched, modelled, and quality-checked end to end by autonomous Claude Fable 5.1 agent swarms, then shipped as plain Three.js apps you can run with npm run dev.

No game engine. No proprietary 3D tiles. Every building, storefront, sign, tree and traffic light is generated from open data and public reference imagery by code that lives in this repo.

Worlds

🌉 Union Square, San Francisco

▶ Watch the full 59-second walkthrough · 1920×1080 · aerial, plaza, Nintendo, lower level, Apple

The square and its surrounding blocks — Powell, Geary, Post and Stockton — on real terrain and a real street grid, with 129 identified storefronts, working traffic lights and cable cars, day/sunset/night, and two explorable interiors: Apple Union Square (300 Post St) and Nintendo SAN FRANCISCO (331 Powell St).


Run it	`cd union-square-sf && npm install && npm run dev`
Buildings	453 OSM footprints · 75 hand-authored façades · 129 named storefronts
Life	220 pedestrians on a 1,398-node nav graph · 109 vehicles incl. Powell St cable cars
Interiors	Apple + Nintendo, 23 interactive objects
Validation	34 camera-matched viewpoints vs. real photographs · 147 comparison sheets · 9 independent reviewer reports

More worlds coming.

How these are made

Each world follows the same pipeline, and every stage is in the repo so you can re-run it:

Reconnaissance — parallel research agents pull OpenStreetMap geometry, USGS elevation, transit and street specs, and a storefront census with per-fact sources and confidence levels.
Offline asset generation — Blender-as-a-library (bpy) scripts emit optimised GLB kits: façade modules, street furniture, vehicles, vegetation, retail fixtures, pedestrian body parts.
Runtime — a pure Three.js app assembles terrain, streets, façades, props, crowds and traffic from JSON specs.
Camera-match QA — Playwright drives the real app, screenshots fixed viewpoints, and diffs them against free-licensed photographs taken from the same spot; independent reviewer agents (architect, geographer, technical artist, interaction) file reports that drive the next fix cycle.

License

Code and generated assets: MIT. Geometry is derived from OpenStreetMap (ODbL) and USGS 3DEP (public domain). Reference photographs are not redistributed here — their provenance is recorded per sector in each world's refs/*/SOURCES.md. Brand names and logos identify the real businesses at their real locations and belong to their owners.