The Shot You Almost Got
You're squeezing four people into one frame at a restaurant table. You angle the phone left, get everyone in, tap the shutter. The photo comes back and the person on the far left looks like they're eating dinner in a different, worse restaurant. Everyone in the centre: crisp, warmly lit. That person at the edge: slightly dark, a little soft, skin tones a shade off. You didn't move. The light didn't change.
So what happened?
Two separate problems, stacked. One is physics. One is software. Together, they reliably punish anyone you push toward the corner.
Light Doesn't Travel in Straight Lines Through Glass
Every lens lets in less light at its edges than at its centre. This isn't a flaw. It's geometry.
When light passes through a lens at a steep angle (the angle needed to reach the corner of a sensor), it travels through more glass, hits the aperture walls at a sharper oblique, and loses intensity on the way. The result is a darkening toward the frame corners called vignetting. In a full-size camera with a large sensor and a well-corrected lens, engineers can spend real money minimising it. In a smartphone, the lens is roughly the diameter of a pencil eraser, the sensor is a few millimetres wide, and the whole stack sits under 8mm deep. Physics doesn't leave much room to negotiate.
The falloff is real: in an uncorrected smartphone lens, light intensity at the extreme corners can be 40 to 60 percent of what it is at the centre. That's not subtle. That's the difference between a face looking properly exposed and looking like it's standing in a closet.
Manufacturers fight back two ways. First, lens coatings and optical element arrangements reduce the worst of it. Second, the image signal processor applies a lens shading correction map, a per-pixel brightness adjustment baked into the camera firmware that lifts the corners back up. On most modern phones this is aggressive enough that you won't notice vignetting in even lighting. But correction maps are tuned for average scenes. When the light is uneven, when there's a backlit window or a dim restaurant, that correction can under-compensate or over-compensate in ways that become visible on a face.
The Metering Problem Nobody Thinks About
The deeper issue isn't optical. Smartphones don't just capture light passively. They decide, in real time, what the "correct" exposure for a scene is, and that decision has a geography.
Every phone's camera app uses a metering algorithm: a system that samples brightness values across the frame and calculates a target exposure. The most common mode, used by default on iPhones, Pixel phones, and Samsung Galaxy devices alike, is centre-weighted or scene-average metering. It pays more attention to what's in the middle of the frame. The logic is sound: for decades, the subject of a photo was expected to be in the centre. Faces especially.
So when you push someone to the edge, the metering algorithm is, in a very literal sense, not looking at them.
It's looking at the table, the background, the jacket of the person standing next to them. If there's a bright wall in the centre, the algorithm darkens the exposure to protect those highlights, and the edge face, already receiving less light due to lens falloff, drops further into shadow. If the centre is dark, it brightens, and the edge face can blow out.
Consider this: two friends, Maya and Daniel, photographed at the same outdoor table in late afternoon sun. Maya is centred in the frame. Daniel is at the left edge, with a bright sky filling the middle of the shot. The phone meters for the sky-heavy centre, pulls exposure down, and Daniel ends up underexposed by roughly a stop and a half. Maya looks great. Daniel looks like he arrived late and the light gave up on him.
The gap between their exposures isn't because they were in different light. It's because the camera chose to care about one of them. That's not a neutral design decision; it's a decades-old assumption baked into firmware, and it still costs people in group photos every single day.
Face Detection Is Supposed to Fix This (It Mostly Does)
Modern smartphones don't only rely on metering zones. They run face detection continuously, using it to bias both focus and exposure toward detected faces regardless of where they sit in the frame. Apple's Smart HDR system, Google's Real Tone processing, and Samsung's scene optimiser all do versions of this. When they work, they work well: the phone spots a face at the edge, locks exposure to protect skin tones, and the corner-physics problem shrinks to a non-issue.
But face detection has real failure modes.
It struggles with faces in profile. It can miss faces that are small relative to the frame, think a wide group shot from five metres away. It gets confused by faces partially occluded by a hand, a glass, or another person's shoulder. In low light, detection confidence drops sharply, and the algorithm falls back on scene metering. In those exact moments, the edge-face problem reappears, and it reappears hardest.
There's also a subtler failure worth knowing about. Face detection finds faces, but it doesn't always weight them equally. In a frame with one centred face and one edge face, some metering systems will average the two exposures, which means the edge face gets half the exposure priority it would have if it were alone in the frame. The centred face effectively outbids it.
The Ultrawide Lens Makes It Worse
When people notice the edge-face problem, the instinct is often to switch to the ultrawide lens to fit everyone in more comfortably.
This is exactly backwards.
Ultrawide lenses (the 0.5x option on most flagship phones, covering roughly 120 degrees) have more extreme angles of incidence at the frame edges than standard lenses. The lens shading correction has more work to do. Geometric distortion is higher. Because the field of view is so broad, subjects at the edge are more likely to fall outside the centre-weighted metering zone entirely. The ultrawide is the worst lens to reach for when you care about edge-face exposure.
The standard lens (1x, typically 24-28mm equivalent) or a slight zoom (2x) is almost always the better call for group shots. It keeps the optical geometry tighter, reduces the metering disparity, and gives face detection a larger, more confidently detected face to lock onto. Two stops of focal length can do more for a group portrait than any post-processing trick.
A Few Things That Actually Help
Tap to expose. On any major smartphone camera app, tapping on a face in the viewfinder before you shoot forces the metering to prioritise that point. If you tap the edge face, the phone recalculates exposure around them. It takes an extra second. Worth it every time.
Use the exposure lock. On iOS, press and hold on a face to lock both focus and exposure (you'll see AE/AF LOCK appear). On Android, the same long-press behaviour exists in the stock camera and most third-party apps. This is especially useful in tricky light where the scene might shift between your tap and your shot.
Shoot in a format that gives you room to fix it. RAW files preserve the full sensor data before the ISP bakes in its exposure decisions. If your phone supports ProRAW (iPhone) or RAW capture (Pixel, some Samsung models), the edge-face problem doesn't disappear optically, but you get far more latitude to recover the shadows in editing without the colour banding you'd get trying to lift a heavily-compressed JPEG.
And recompose when you can. If one person consistently ends up at the edge, take two steps back. More distance means a smaller angular difference between the centred subject and the edge subject, the lens falloff eases, the metering zone covers more of the frame, and the physics problem gets smaller just by changing where you stand.
The Corner Is Not a Safe Place
The edge-face problem reveals a quiet assumption built into camera design: the subject belongs in the middle. Decades of photography conventions, lens engineering tradeoffs, and metering logic all quietly conspire against anyone you push to the side of a frame.
Face detection has made real progress against this. But it's a software patch over a physics problem, and software patches have conditions.
When the light gets difficult, when the face gets small, when the angle gets steep, the patch peels back and the physics wins. Ask yourself whether any photo you've dismissed as "bad lighting" was actually just bad placement.
Knowing this won't stop you from composing the shot that puts someone at the edge. But it should stop you from blaming the light.