Deep Fusion will take nine images of varying exposure lengths for every single click of the camera’s shutter a user makes. Eight of these will be taken before the user even clicks the shutter button, followed by a ninth long exposure when the user presses the shutter button. Then, in one second, Deep Fusion will go through the captured 24 million pixels to pick and choose the most detailed and lowest noise ones to stitch together a perfect Franken-photo. “This is the first time a Neural Engine is responsible for generating the output image,” Schiller said onstage.