Analysis updated 2026-05-18
Explore a procedurally generated 4 by 4 kilometre landscape with a day and night cycle in your browser.
Study a real example of an AI model designing and writing nearly all the code for a large graphics project.
Use the preset viewpoints or automated flythrough to see composed scenic locations without manual navigation.
| braffolk/fable5-world-demo | adrienckr/notslop | alchemz/solana-pumpfun-token-bundler | |
|---|---|---|---|
| Stars | 78 | 78 | 78 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | easy | hard |
| Complexity | 3/5 | 2/5 | 4/5 |
| Audience | developer | writer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a recent desktop Chrome browser with WebGPU support, no mobile support.
LAAS is a procedural 3D open world that runs inside a web browser, built almost entirely by an AI model called Claude Fable 5. The repository contains no pre-made images, 3D models, or audio files. Every tree, rock, river, cloud, and blade of grass is generated by code each time the world loads. The world covers a 4 by 4 kilometre landscape with mountains, forests, rivers, lakes, meadows, and a full day and night cycle with dynamic weather. Around 190,000 individually unique trees are placed across the terrain, alongside hundreds of thousands of smaller plants, and roughly a million grass blades render across the meadows. The rendering engine uses a newer browser graphics technology called WebGPU, which means the page currently only works in recent versions of Chrome on a desktop or laptop computer. No mobile support is available. To run it you clone the repository, install dependencies with npm, and open a local URL in Chrome. You can walk, sprint, jump, and fly around the world using keyboard and mouse controls. A set of preset viewpoints lets you jump directly to composed scenic locations, and a 90-second automated flythrough is also included. The project was created as a test of what Claude Fable 5 could build with minimal human direction. The human contributor wrote one planning document describing the visual goals and hard constraints, then let the model design the architecture and write nearly all the code. The model also ran its own quality checks by booting the world headlessly in a browser, taking screenshots, comparing frames against reference images, and profiling the graphics card workload. The codebase is around 21,000 lines of TypeScript, with a running log of decisions, measurements, and known issues kept in a file called STATUS.md that the model maintained as its own working memory between sessions.
A procedurally generated 3D open world you explore in Chrome, built almost entirely by an AI model.
Mainly TypeScript. The stack also includes TypeScript, WebGPU, npm.
No license information is stated in the source material.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.