it's much easier to record it without much shake and then maximize the effect of their algorithm after the fact instead of recording a bunch of versions with varying degrees of shake. They completely overdid it in my opinion, the example doesn't seem authentic at all to me (though the technology remains cool)
