New AI Could Fix the Blown-Out Photos on Your Phone

Have you ever photographed a sunset and gotten a white blob where the sun should be? That's a dynamic range problem. A camera can only capture so much brightness at once, so it has to choose: keep the sky pretty and let the buildings go black, or keep the buildings visible and let the sky turn white. "HDR" (high dynamic range) means capturing both at the same time. A team of researchers has now published a new method for doing that, accepted at SIGGRAPH Asia 2026, a major graphics conference.

Their trick is to dodge the hard problem. Instead of asking one AI to rebuild an entire scene's full range of light in a single leap, they train it to do something much easier: take a picture and brighten its highlights by one "exposure value" — roughly one click of a camera's brightness dial. Then they run that same small step over and over, each round pulling a bit more detail out of the blown-out areas. Learn one easy move, repeat it. That approach let them train on ordinary RAW photos, the unprocessed files your phone already saves.

For anyone who takes pictures, the payoff is real. Think candlelit dinners, neon night streets, a bright window indoors, stage lights at a concert — moments where your phone gives up and produces a glowing white haze. The system is especially good at the hard cases: bright light sources and highlights, which is exactly where most automatic fixes fail.

Two honest caveats. First, this is a research paper, not an app. Nothing lands on your camera tomorrow, though photo software could eventually adopt it. Second, the AI is guessing: it invents plausible detail in areas your camera never clearly recorded. Sometimes that looks magical. Sometimes it will simply be wrong.