# World-GAN: a Generative Model for Minecraft Worlds

**Authors:** Maren Awiszus, Frederik Schubert, Bodo Rosenhahn  
**arXiv:** [2106.10155](https://arxiv.org/abs/2106.10155)  
**Date:** 2021-06-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2106.10155v1/

## Key points

- World-GAN is a method for data-driven Procedural Content Generation via Machine Learning in Minecraft from a single example.
- It is based on a 3D Generative Adversarial Network (GAN) architecture that can create arbitrarily sized world snippets from a given sample.
- The proposed block2vec representations make World-GAN independent of the number of different blocks which can vary significantly in Minecraft.
- This feature enables the generation of larger levels and allows users to generate Minecraft worlds based on parts of their creations.
- World-GAN also demonstrates that changing this new representation space allows us to change the generated style of an already trained generator, providing users with even more control over their generated content.
- Its ability to create complex worlds from just one example makes it an invaluable tool for game developers and enthusiasts alike.

## Summary

World-GAN is a groundbreaking method for data-driven Procedural Content Generation via Machine Learning in Minecraft from a single example. Developed by Maren Awiszus, Frederik Schubert, and Bodo Rosenhahn, this approach is based on a 3D Generative Adversarial Network (GAN) architecture that can create arbitrarily sized world snippets from a given sample. The authors evaluated their approach on creations from the community as well as structures generated with the Minecraft World Generator. The proposed block2vec representations make World-GAN independent of the number of different blocks which can vary significantly in Minecraft. This feature enables the generation of larger levels and allows users to generate Minecraft worlds based on parts of their creations. The authors were motivated by the dense representations used in Natural Language Processing (NLP) introduced with word2vec [1]. In addition to its impressive capabilities, World-GAN also demonstrates that changing this new representation space allows us to change the generated style of an already trained generator. This feature provides users with even more control over their generated content. Overall, World-GAN represents a significant breakthrough in machine learning-based procedural content generation for Minecraft. Its ability to create complex worlds from just one example makes it an invaluable tool for game developers and enthusiasts alike.