When you press the shutter, a modern smartphone has already been watching. It captures a stream of frames, picks the sharpest, aligns them and averages away noise. The result looks like one photograph, but it is a software composite: computational photography in action.
- Bursts, not single shots: Phones merge several frames to suppress noise and recover detail.
- HDR+ uses short, dark exposures: Averaging protects highlights, then shadows are lifted with less grain.
- Night Sight and Deep Fusion add AI: Longer frames, learned color and pixel-level processing extend what small optics can do.
The camera starts before you press the shutter
Google’s Pixel phones use zero-shutter-lag capture. As soon as the camera app opens, frames enter a circular buffer that discards older images. When you tap the shutter, the software receives the 9 or 15 most recent frames. That is why the photo can match the instant you intended. Because those frames also feed the viewfinder, HDR+ caps each exposure at 66 milliseconds so the display stays at 15 frames per second or more.
One frame is rarely enough. A phone lens is small and its pixels are tiny, producing noise in low light and limiting dynamic range in high-contrast scenes. Longer exposures invite blur because hands are never perfectly still. The sensor also does not see full color: each pixel measures only red, green or blue through a Bayer mosaic, and the rest is inferred. Google has noted that two-thirds of a digital RGB image is reconstruction. DSLRs face the same issue, but phones have less optical margin.
HDR+ and the art of averaging
Classic HDR blends frames taken at different exposures. Google’s original HDR+, described in 2014 for the Nexus 5 and Nexus 6, captured a burst at one short, deliberately dark exposure to protect highlights. It aligned and averaged the frames: random noise canceled, while the repeated signal grew stronger. Cleaner shadows could then be brightened without obvious grain. Similar frames aligned reliably, avoiding the ghosts and double edges of traditional HDR.
The sharpest frame served as a reference, a technique astronomers call lucky imaging. Capture lasted one-third of a second to one second. By 2021 Google acknowledged a limit: with every frame underexposed, shadows in contrasty scenes still carried noise. The fix, on the Pixel 5 and Pixel 4a 5G, was HDR+ with bracketing, which mixed exposures too. Google’s HDR+ became a hybrid.
Night Sight, Deep Fusion and zoom from handshake
Night Sight pushes further: fewer frames, much longer exposures and machine learning for color. Google’s figures show the progression. A single-shot phone struggles at 30 lux; HDR+ held up to 3 lux; Night Sight targets 3 to 0.3 lux without a tripod or flash. Before shooting, the phone measures scene and hand movement, then adjusts exposure. With little motion, each frame can last up to 333 milliseconds. A handheld sky example used 15 frames of 333 ms, about five seconds; on a tripod, six one-second frames.
Color is another challenge. In very low light, white balance fails, so Google trained an algorithm on hand-corrected scenes. It also made an aesthetic choice: raise contrast and push shadows toward black, as painters did, so the photo still reads as night. Apple entered the field with Night mode on the iPhone 11 Pro in 2019.
Apple promised better indoor and outdoor low-light shots with “natural colors and less noise.”
Apple also introduced Deep Fusion, powered by the Neural Engine in the A13 Bionic, which processes images pixel by pixel to optimize texture, detail and noise. Photonic Engine followed with the iPhone 14 Pro in 2022, applying Deep Fusion earlier. Apple said it improved mid- and low-light shots by up to 2x on the main camera and 3x on the ultrawide.
Super Res Zoom, introduced by Google in 2018, exploits hand tremor. Stabilization cannot remove every tiny shift, so the image moves a few pixels between frames. Software uses those shifts to recover information a single shot lacks and reconstruct more detail when zooming.
What this means when you take a photo
The practical lesson: if the phone fuses a burst, hold still after pressing the shutter. Google asked users in 2014 not to move until the capture animation finished. For fast subjects such as sports, its engineers disabled HDR+ because multi-frame capture takes longer than a single shot.
The interesting question is no longer whether a phone “retouches” a photo; it always has. It is how much the device decides. Averaging bursts to remove noise is applied physics. Choosing that night should look like night, or that skin should have a particular texture, is a manufacturer’s aesthetic judgment. The boundary between reconstructing detail and inventing it is thin. A phone photo remains a record of what happened, but an interpreted one.