In this paper we present a new approach for real time cartoon-style rendering for large datasets. It uses color quantization to determine the different shadow regions and utilizes multiresolution hierarchy to maintain appropriate level of detail over the various regions of the mesh. To comply with visual requirements, our algorithm performs real time mesh subdivision on selected regions to reach the appropriate level of detail. In addition, we have adopted vertex split and edge collapse operations that can be performed without incurring extra vertex split/edge collapse operations as a precondition. The real time refinement, which can exceed the original resolution of the mesh, dramatically improves the quality of rendered image while using only a small processing overhead. This refinement can be applied recursively, potentially providing a smooth model surface and smooth cartoon features in any possible camera zoom.