Also, his facial expressions do not, in any way, mesh with the words that he's saying. To say nothing of his gestures. IS THIS THE BEST THE ROBOTS CAN DO?!
think you'd need to preprocess the mesh and encode the per vertex ratio of uv density to spatial density in say a vertex alpha channel. not sure on the best method to compute this but off the top of my head I'd say.... <area of all faces used by the vert >/<area of all map faces used by the vert>