Since the area the generated image changes is known between frames, the animations could be made more consistent by only changing the relevant masked pixels instead of the full images.