xandergos/terrain-diffusion-mc
Procedural terrain generation with diffusion models (in Minecraft)
What it solves
This project replaces standard Minecraft world generation with a system based on a diffusion model. It solves the problem of creating more diverse and realistic terrain that goes beyond the traditional procedural generation methods used in the game.
How it works
The mod integrates a research-backed diffusion model into the Minecraft Fabric environment. It generates elevation and four climate variables, which are then converted into Minecraft biomes using a set of hand-written rules. The system supports hardware acceleration via DirectML (Windows), CUDA (NVIDIA GPUs), and CoreML (Apple Silicon) to handle the computationally intensive inference process.
Who it’s for
Minecraft players who want more realistic and diverse world landscapes, as well as mod developers interested in improving biome classification for AI-generated terrain.
Highlights
- Research-backed: Based on research accepted to SIGGRAPH 2026.
- Hardware Acceleration: Supports multiple GPU backends for faster terrain generation.
- Terrain Explorer: Includes a built-in web UI that allows players to scout continents, mountains, and islands via an interactive map before exploring them in-game.
- Customizable Scale: Allows users to adjust the world scale to balance the level of detail and hardware performance.