Terrain and rivers
Height comes from layered noise, then hydraulic erosion carves the ridges and gorges, then river channels are laid along the flow field. One seed always gives the same world; a new seed gives a new continent.
World Lab is an experiment in real-time procedural generation. The ridges, gorges, forests and cloud layers you see are not authored models — your GPU works them out the moment you press Enter World. The whole world downloads in about 1.1 MB.
Requires desktop Chrome with WebGPU. No sign-up.
Height comes from layered noise, then hydraulic erosion carves the ridges and gorges, then river channels are laid along the flow field. One seed always gives the same world; a new seed gives a new continent.
Trees are grown, not placed: a skeleton first, then bark synthesised at runtime and leaves built on top. Distant stands collapse to impostors; up close there is ground cover, understory and deadfall, scattered by slope and moisture.
Volumetric cloud layers, froxel atmospheric scattering, cascaded shadows, screen-space ambient occlusion and probe GI. Scrub the time of day from dawn to night and the whole lighting solution follows.
Walk or fly, no level streaming, no loading screens. Photo mode, shareable view links and exploration objectives are built in; F3 shows live draw calls and triangle counts.
This was a test of where WebGPU tops out. If everything in a world can be computed at runtime, then loading stops existing as a problem — no models to download, no textures to decode, no CDN bill. It turns out you can, and the price is that the entire budget moves onto the GPU. The value is not in making games; it is in anything that needs space that is unbounded, reproducible and weightless: digital-twin base layers, immersive product showcases, procedural asset production.
World Lab is open source. If you are building a digital twin, a 3D visualisation, or anything that needs generation rather than authoring, let us talk about how it lands in your case.
Talk about your project