Virtual worlds are ubiquitous in video games, films, or medical and industrial applications today. Until now, creating realistic 3D content has been very time-consuming and labor-intensive. Previous methods for automating the process do not come close to the quality of the work of artists who design virtual worlds by manual labor. MDSI Core Member Matthias Nießner, Professor of Visual Computing, is tackling this challenge in his project "Gen3D: Learning to Create Virtual Worlds". He develops generative models that generate 3D polygon meshes and their surface textures and material properties. These models can be used directly by modern graphics systems. He uses images and videos as training data. As these only partially show the 3D world, the models must also learn to deal with incomplete data.