Simple Face Tracker with Smoothing

Developed a way to memorize information between chunks of video generated using a video diffusion model using point-clouds. Using a fast rendering approach, we update the latent chunks to allow for long range video generation without error accumulation.