KM uses HTML5 Canvas to render the animation.
One way would be to draw the current video frame like an image.
The other way would be to put a video object on top of the canvas as a layer.
The first option probably is more flexible but most likely will also use more cpu power and play the video less fluently. The second option most likely will give the best video quality but the disadvantage is that the video object always is on top and nothing can be put over it.