This wouldn't seem to help, unless you correlate it with motion sensing data. You can build point clouds from deep fake videos just as easily as from real videos.