In an image processing system, a method for generating a images includes rendering graphical models comprising a scene to separate image layers called "gsprites," and then compositing these image layers to generate an image. An image processor can retrieve gsprites from memory, transform them, and composite them for display at video rates. Gsprites can be re-rendered or updated at different rates. Reducing the rendering overhead of the system, the image processor can perform an affine transformation on the gsprite to approximate motion of the graphical object that it represents, rather than re-render the object. Objects in a scene can be queued for re-rendering based on a predefined update rate, or based on the accuracy of representing the object with a transformed gsprite, rendered for a previously displayed image.