We describe our approach to detecting fake biometrics displayed on a modern smartphone screen. Using our dataset of around 100k images we have trained neural networks to differentiate a recaptured image from an authentic one, achieving over 99% accuracy.