In this thesis, we address the problem of automatic generation of graphical content with a high level of detail for the generation of worlds. In this quest for realism, the major scientific and technical challenges include: managing the mass of geometric data needed to create varieties of natural objects, taking into account interactions with the various objects and user control.Our approach is based on an original volumetric model for representing and unifying the different materials of land. We offer a range of high-level algorithms, combining simulation steps to ensure physical plausibility and visual realism, and procedural generation techniques to effectively manage complexity. Our algorithms are used to sculpt and complex terrain with caves and overhangs and generate roads with tunnels and bridges to cross natural barriers.
