In my last talk, I touched upon the concept of projective reconstruction and its computation and use in shape-from-stereo. In this talk, I will expand the idea to multiple non-parallel views, and show the result of a novel algorithm using matrix factorization and weighted least squares. This lays a computational framework for IBRM. Two applications will be demonstrated: view synthesis and object insertion. It is hoped that more multimedia applications such as face modeling/animation, virtual/augmented reality and 3-D video editing can be built on top of this research.