PhiloLabs/fable51-worlds

What it solves

This project creates walkable, 3D virtual environments based on simple prompts (text, images, or video). It replaces traditional manual 3D modeling or pixel-based AI generation with a "worlds as code" approach, allowing for precise, verifiable, and editable reconstructions of real-world locations in a web browser.

How it works

The system uses a pipeline of AI agent swarms to transform a brief into a Three.js application:

1. Reconnaissance: Research agents gather map geometry, elevation data, and street specifications.
2. Asset Generation: Scripts create the necessary 3D components (buildings, street furniture, vegetation) as code or JSON specs.
3. Runtime: A Three.js app assembles these components into a walkable scene.
4. Verification: The system uses Playwright to take screenshots and compare them against real photographs, with independent reviewer agents (architects, geographers) filing reports to fix defects.

Who it’s for

Developers, technical artists, and researchers interested in synthetic environments, digital twins, and agent-based world-building for animation, robotics simulation, or RPGs.

Highlights

• Code-based Generation: Every element is generated by code rather than proprietary 3D tiles or downloaded meshes, making the world fully editable via diffs.
• Multi-modal Input: Can generate worlds from a single sentence, a photograph, or a video clip.
• Verifiable Accuracy: Uses automated diffing against real-world photos to ensure geometric and visual fidelity.
• No Game Engine Required: Ships as a lightweight Three.js app that runs directly in the browser.