Abstract: Cloud simulation models are rare in computer graphics, although many rendering algorithms have been developed to evaluate the illumination and the color of gaseous phenomena. The laws of fluid mechanics used for physical simulation require a fine resolution in space and time, and solving the Navier-Stokes equation in 3D is in general quite costly. However, many heuristics, dealing with various scales, can be used to describe the evolution of the shape of convective clouds such as cumulus. These go beyond the classical equations governing the motion of each element of fluid volume. Physicists characterize the identity and behavior of phenomena such as bubbles, jets, instabilities, waves, convective cells, and vortices. Moreover, the shape of a convective cloud can be considered as a surface, so that we only require detailed information near the periphery of the cloud volume. A guiding principle in our ongoing research is to take advantage of this type of high level knowledge available at various scales in order to obtain a simulation of convective clouds that may not be physically-accurate, but that will be perceptually convincing.