Our extensions to multidimensional lightcuts improve rendering performance using precomputation and hierarchial approaches. The first approach precomputes static illumination of static geometry, but defers the remaining light path computations until render time. Once the dynamic content of a scene is known, the remaining light paths are resolved and a final image is computed. The impact of precomputation impacts performance in a view and scene-dependent manner. We also explore a hierarchial approach that computes coarse regional approximations for lighting, then progressively refines these approximations to an error-bounded threshold as necessary. Performance improvements for this hybrid method depend on the realization of fast visibility-testing, either through the use of hardware shadow mapping or packetized ray tracers.