The goal of the project is to propose high-level techniques to assist artists in authoring and creating virtual worlds. This help will be provided as high-level tools that will support users in their tasks, without introducing any trade-off in the creative pipeline. This means that artists should not be limited by the introduction of such tools, and particularly, they should keep different levels of control. The project will rely on machine learning methods to predict detailed maps from sketches or coarse input maps.