How does a phone turn a dark scene into a bright photo?

Your phone does not take one night photo. It takes a burst of them and merges them, because the problem in the dark is a counting problem. Image noise comes largely from shot noise, the natural variation in the number of photons entering the lens, which matters most for smartphone cameras, with their small lenses and sensors. A second source is read noise: random errors when each pixel's charge is converted to a number. The signal-to-noise ratio rises with the square root of the exposure time or faster, so a longer exposure gives a cleaner picture. But it is hard to hold still in dim light, and whatever you are photographing probably will not hold still either.

Google's answer, HDR+, introduced in 2014, captures a burst of frames, aligns them in software and merges them. Its main purpose is dynamic range, but merging also reduces the impact of shot and read noise. Each frame uses a short exposure to stay sharp, and pieces of frames that cannot be aligned are rejected. The idea of averaging frames is as old as digital imaging: astrophotographers call it exposure stacking, and the hard part is alignment when the camera is handheld.

Night Sight pushes this further. Single-picture phones struggle at 30 lux, merging phones do well down to 3 lux, and Night Sight aimed at the range between 3 and 0.3 lux with no flash. A handheld sky shot took 15 frames of 333 ms each, 5 seconds of capture; on a tripod, 6 frames of 1 second.

The result is not a record of what your eyes saw. In very dim light humans stop seeing in colour, because the cone cells stop functioning, leaving rod cells. Night Sight makes pictures brighter than reality, which its own authors call a "look".