A system accesses a virtual scene including a three-dimensional (3D) object and a mesh object that models a cloth object. The system performs a refinement simulation to model a drape of the cloth object over the 3D object. Performing the refinement simulation includes, for each of a sequence of mesh resolutions: determining a configuration of the mesh model that minimizes a proxy energy function of a finest mesh resolution of the sequence of mesh resolutions. The system generates, for display via a user interface during the refinement simulation, an editable preview object comprising the mesh object at a coarsest level mesh resolution. The system receives a modification to the editable preview object and displays the modified editable preview object. A configuration of a finest level mesh resolution of the mesh object in the refinement simulation is geometrically consistent with a configuration of the modified editable preview object.