Techniques provided herein apply a precomputed graphical object to one or more images to generate a video that is modified with the precomputed graphical object. Various implementations characterize facial positions on a face in a first image and determine a respective facial position on the face to apply a precomputed graphical object at. One or more implementations modify the first image by applying the precomputed graphical object to the respective facial position in the first image. Some implementations modify one or more images that are captured after the first image by applying the precomputed graphical object to each respective location for the respective facial position in the one or more images. In turn, various implementations generate a video with images that are modified based on the precomputed graphical object.