AI/TLDR

Two Minute Papers · 2026-07-12 · notable

Two Minute Papers: 'Minecraft Was Missing One Brilliant Idea'

Two Minute Papers walks through InfiniteDiffusion by Alexander Goslin — a training-free algorithm that turns any diffusion model into an infinite, deterministic terrain generator, shipped as a Minecraft mod and a Unity demo.

Two Minute Papers thumbnail for 'Minecraft Was Missing One Brilliant Idea'.

A Two Minute Papers walkthrough of InfiniteDiffusion — a diffusion algorithm that generates unbounded terrain in real time.

What is it?

Two Minute Papers' July 12 video reviews InfiniteDiffusion, a SIGGRAPH 2026 paper by independent researcher Alexander Goslin. The method is training-free: it wraps any pre-trained diffusion model into a seed-consistent, infinitely tileable terrain generator. The paper ships alongside an MIT-licensed GitHub repo and a Minecraft Fabric mod on Modrinth.

How does it work?

InfiniteDiffusion combines a hierarchical diffusion stack with a Laplacian encoding so a single model can span the full ~-10,000 m to ~9,000 m Earth elevation range. An open-source infinite-tensor framework gives O(1) random access to any tile, so the same seed always produces the same landscape at any coordinate — no border artifacts, no re-generation.

Why does it matter?

The video pitches diffusion as a practical successor to Perlin noise for game worlds and simulations: random-access, seed-consistent, and running interactively on a consumer GPU with ~1.5 GB of VRAM. Karoly's channel is a common first-look explainer for engineers tracking practical applications of diffusion beyond image generation.

Who is it for?

Game developers, procedural-graphics researchers, and diffusion-model engineers.

Try it

Watch on YouTube: https://www.youtube.com/watch?v=Ae9q7KsRbuI

Sources · 4 outlets

Tags

  • two-minute-papers
  • youtube
  • video
  • infinite-diffusion
  • terrain-diffusion
  • diffusion-models
  • procedural-generation
  • siggraph-2026
  • minecraft
  • computer-graphics

← All releases · Learn AI