Apple is actively testing artificial intelligence to develop computational photography capabilities. One of the key projects is the DarkDiff neural network, designed to denoise and preserve details in night images with minimal loss of quality. This was reported by 9to5Mac.
Unlike traditional solutions, the new system does not process the finished image but works directly with noisy areas during frame saving. The algorithm is based on the open V2-1 Steady Diffusion model.
At the current stage, development is in the initial testing phase. DarkDiff still consumes too many resources to run in the background on mobile devices and also does not correctly interpret non-Latin characters on symbols or text, which can lead to artifacts.
In parallel, Apple is secretly working on its own CMOS sensor with a similar built-in noise reduction feature, which is currently being tested for future iPhones. It is expected that bringing these technologies into mass production may take several years.
