Fable 5.1 Agent 集群生成可交互3D世界开源项目

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).

⛩️ Higashiyama, Kyoto

▶ Watch the full 54-second walkthrough · 1920×1080 · seven scenes — over the roofs, the Shirakawa, Yasaka Shrine, the weeping cherry, the Yasaka Pagoda, Sannenzaka, and Kiyomizu-dera at sunset

Kyoto's Southern Higashiyama walking route — Gion → Hanamikoji → Yasaka Shrine → Nene-no-michi → the Yasaka Pagoda → Ninenzaka → Sannenzaka → Kiyomizu-zaka → Kiyomizu-dera — 2.3 km of it, continuously walkable, climbing 76 m from the ochaya on Hanamikoji to the temple stage. Rendered 3D-to-2D as a hand-painted anime background: cel materials with hue-shifted shadow bands, screen-space ink taken from a second difference of depth, and a split-tone grade. No binary assets at all — every sign, noren, lantern, roof tile and paving stone is drawn with Canvas2D at start-up.

Six widely-repeated figures were overturned by the survey and the world is built on the measured ones — the Yasaka Pagoda is 38.79 m, not 46, with a convex taper; the Kiyomizu stage deck is at 115.5 m ASL, not 240; the stage is 21.8 × 9.6 m on 168 pillars of 0.64 m diameter, not 18 × 10 m on 139 pillars of 2 m.

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.