Samsung Patents a Two-AI System for Merging Multiple Camera Frames Into One Sharp Photo

By Patentlyze Team · Updated Jul 21, 2026

Taking a photo in tricky light means your camera is gambling: expose for the bright parts and lose the shadows, or expose for the shadows and blow out the highlights. Samsung's new patent describes a system that uses two separate AI models to handle both sides of that tradeoff at once.

How Samsung's dual-AI photo system handles HDR shots

Imagine you're photographing a friend standing in front of a sunny window. The window looks white and washed out, or your friend looks like a silhouette. Your camera can't easily nail both at the same time. Samsung's patent addresses this by having the phone snap multiple shots, including some that are deliberately inverted (think of a photo negative from old film days), each at different brightness levels.

One AI model processes the normal frames and figures out which parts of each shot are in motion, so it doesn't blur together a hand that moved between shots. A second AI model takes the inverted frames and reconstructs color information from them. Both outputs get combined into a single high dynamic range (HDR) image that holds detail across bright and dark areas at the same time.

The final step is "tone mapping," which is basically squishing that wide-range HDR result down into a version your phone screen can actually display without looking washed out or artificially processed. The whole pipeline is designed to run on the phone itself, not in a cloud server.

How the two AI models split and process each burst

The patent describes a pipeline with two parallel processing tracks that eventually merge into one output image.

Track 1 (normal frames): The camera captures multiple shots at a standard exposure. Each frame is converted into two color formats, YUV (a format that separates brightness from color, useful for detecting motion) and RGB (standard red-green-blue color data). Motion maps are built from the YUV frames to flag areas where something moved between shots. A first AI model then uses those motion maps alongside aligned color data to produce a single clean composite frame.

Track 2 (negative frames): The camera also captures "negative" frames at several different exposure levels. A negative image is essentially an inverted exposure, capturing tonal information that the normal frames miss. These negatives are aligned and blended into one color filter array image (the raw grid of color data a sensor captures before it becomes a viewable photo). A second AI model converts that into a usable color image.

The two tracks are then blended together into an HDR image, which is put through tone mapping (a compression step that makes the wide tonal range fit a standard display) to produce the final photo. The use of two specialized AI models, one focused on motion and alignment, the other focused on reconstructing color from inverted data, is the core engineering choice here.

What this means for Samsung's night and HDR photography

HDR photography on phones has existed for years, but most implementations blend a small number of bracketed exposures using relatively simple algorithms. Samsung's approach adds negative frames as a separate input source and dedicates an entire AI model specifically to extracting useful color data from them. That is a meaningful architectural difference from the single-model pipelines most phones use today.

For Galaxy users, the practical payoff would be photos where both the sunlit sky and the shaded foreground look properly exposed, with less of the artificial or "HDR-ified" look that older processing creates. It also signals Samsung is investing in on-device AI camera processing to compete with Google's Pixel computational photography stack.

Editorial commentary on a publicly published patent application. Not legal advice.