PhiloLabs/fable51-worlds

worlds via code, from fable 5.1

What it solves

This project creates immersive, walkable 3D environments from simple inputs like a text prompt, a photograph, or a video clip. It replaces traditional 3D modeling and game engines with a "worlds as code" approach, allowing for precise, verifiable, and editable 3D scenes that run directly in a web browser via Three.js.

How it works

The system uses a pipeline of AI agent swarms to transform a brief into a functional 3D app:

  1. Reconnaissance: Parallel research agents gather map geometry, elevation data, and street specifications from sources like OpenStreetMap and USGS.
  2. Asset Generation: Scripts generate the necessary 3D assets (façades, vegetation, vehicles) procedurally, often drawing textures at startup rather than using binary files.
  3. Runtime: A Three.js application assembles these components based on JSON specifications.
  4. Verification: The system uses Playwright to take screenshots of the world and compare them against real photographs, with independent reviewer agents (architects, geographers) providing feedback for iterative fixes.

Who it’s for

Developers and creators who want to generate high-fidelity, spatially accurate 3D worlds from minimal input without needing proprietary 3D software or heavy game engines.

Highlights

  • Input Flexibility: Supports text, video, and image inputs to define the subject and style.
  • World-as-Code: Environments are generated as code, making them fully editable via diffs rather than requiring a complete re-roll of the AI generation.
  • No Game Engine: Ships as a plain Three.js app that opens in the browser.
  • Verifiable Accuracy: Uses automated diffing against real-world photographs to ensure geometric and visual fidelity.
  • Agentic Workflow: Employs a swarm of specialized agents for research, building, and QA.

Related

  • Project
    scottstts/Threejs-Awesome-Graphics-Agent-Skills

    Threejs‑Awesome‑Graphics‑Agent‑Skills is an npm package that bundles ready‑to‑use, high‑quality Three.js code snippets (called “skills”) for advanced visual effects—procedural materials, realistic oceans, volumetric clouds, ray‑marched space scenes, and more. It installs these snippets into the directory layout expected by AI coding assistants (Codex, Claude Code, Cursor, etc.), letting the agents copy‑paste proven implementations when a user asks for “awesome graphics”. The repo includes an example gallery, validation scripts, and a deterministic capture pipeline, making it a practical resource for developers and AI‑augmented workflows.

  • Project
    MirroS-Lab/Code-as-World

    Code-as-World is a framework that represents the physical world as executable code to enable AI models to perform quantitative physical reasoning by discovering world representations through iterative simulation.

  • Project
    liyupi/yu-ai-code-mother

    An enterprise-grade zero-code AI application generation platform that transforms text requirements into deployable software using AI agents and workflows.

  • Project
    OpenDCAI/GameFactory-3A

    An open-source framework that uses coding agents to transform game requirements into production-ready assets and engine-specific code for UE5, Unity, Godot, Blender, and three.js.

  • Project
    HorizonRobotics/EmbodiedGen

    A generative 3D world engine that converts text and images into simulation-ready 3D assets and environments for embodied AI, deployable across multiple robotics simulators.