@plasius/gpu-world-generator

GPU-assisted world generation focused on hex-grid terrain synthesis. This package targets pre-generation of terrain height/heat/biome layers using WebGPU compute jobs (compatible with @plasius/gpu-worker).

Goals

• Hierarchical hex grid (1000 km² regional zones down to ~10 m² tiles).
• 3D heat-map based terrain synthesis (height = depth, heat = biome driver).
• Extensible biome classification (tundra, savanna, river, city, village, ice, snow, mountainous, volcanic, road, town, castle, etc).
• Shader-first pipeline with CPU fallback helpers.

Layered Fractal Model

Generation now uses three explicit fractal layers:

• Layer 1 (terrain trend): cumulative height banding where 0.0..0.2 is downward slope, 0.2..0.8 is flat, and 0.8..1.0 is upward slope.
• Layer 2 (features/obstacles): dedicated fractal mask for obstacles and prop placement (rocks, boulders, ruins, water ripples, etc).
• Layer 3 (foliage): dedicated fractal mask for vegetation density (trees, bushes, grass tufts, reeds).

TerrainCell outputs may include surface, feature, obstacle, foliage, and slopeBand in addition to height, heat, moisture, and biome.

Install (local)

npm install
npm run build

Usage (WGSL)

import {
  assembleWorkerWgsl,
  loadJobWgsl,
} from "@plasius/gpu-worker";
import { terrainWgslUrl, loadTerrainWgsl } from "@plasius/gpu-world-generator";

const jobWgsl = await loadTerrainWgsl();
await loadJobWgsl({ wgsl: jobWgsl, label: "terrain" });
const workerWgsl = await loadWorkerWgsl();
const shaderCode = await assembleWorkerWgsl(workerWgsl, { debug: true });

Usage (Temperate Mixed Forest)

import { generateTemperateMixedForest } from "@plasius/gpu-world-generator";

const { levelSpec, cells, terrain } = generateTemperateMixedForest({
  seed: 1337,
  radius: 6,
});

Usage (raw import with bundlers)

import terrainWgsl from "@plasius/gpu-world-generator/terrain.wgsl?raw";

Worker DAG Manifests

@plasius/gpu-world-generator now publishes worker-first generation manifests so chunk and voxel work can be scheduled as a multi-root DAG instead of a flat queue.

import { getWorldGeneratorWorkerManifest } from "@plasius/gpu-world-generator";

const streaming = getWorldGeneratorWorkerManifest();
const bake = getWorldGeneratorWorkerManifest("bake");

console.log(streaming.jobs.map((job) => job.worker.jobType));
console.log(bake.jobs.find((job) => job.key === "assetSerialize"));

• streaming models runtime chunk generation and mesh materialization.
• bake extends the DAG with asset serialization for background/offline output.
• Jobs include queue class, priority, dependencies, adaptive budget ladders, and debug allocation tags for integration with @plasius/gpu-performance and @plasius/gpu-debug.

Render Representation Plans

@plasius/gpu-world-generator now also publishes explicit chunk representation-tier plans so renderer and worker packages can coordinate near, mid, far, and horizon outputs without guessing from distance alone.

import { createWorldGeneratorRepresentationPlan } from "@plasius/gpu-world-generator";

const plan = createWorldGeneratorRepresentationPlan({
  chunkId: "hex-12-9",
  profile: "streaming",
  gameplayImportance: "critical",
});

console.log(plan.bands);
console.log(plan.representations.find((entry) => entry.output === "rtProxy"));

Each plan exposes raster-facing and RT-facing outputs separately, plus refresh cadence, shadow relevance, chunk-identity preservation, and scheduling metadata that downstream renderer and worker packages can prioritize by band and importance.

Demo

The WebGPU mixed-forest demo lives in demo/. Run it with:

cd demo
npm install
npm run dev

Development Checks

npm run lint
npm run typecheck
npm run test:coverage
npm run build
npm run pack:check

Notes

• For Vite/Pnpm setups, raw WGSL import is the most reliable.
• See docs/plan.md for hierarchy and biome rules.
• See docs/design/worker-manifest-integration.md for the chunk/voxel DAG contract.