Games Engineering Programs

GAN for adaptive Terrain Generation

GE GameLab II
Created by
Tobias Böhler, Jan Hoffmann


GANs are a method of content generation via deep learning, composed of two neural networks. One of them is generating (generator) an output trying to mimic a real counterpart. Those outputs are then given to the other network (discriminator), which tries to differentiate between original content and fabricated outputs of the generator. 

In this project we utilized GANs to generate detailed heightmaps from simple scetches, which can then be rendered as a 3d environment. This alows us to create large creative terrains with low effort.

Legal Information