GoKawiilWith iOS 27, Apple is introducing a new version of its system-level RAW image processing engine. It uses machine learning to greatly improve detail and reduce noise, including when reprocessing older RAW photos. Here are the details.
iOS 27 to include RAW 9
If you’re not familiar with RAW, it is basically an image format that preserves the data captured directly by a camera’s sensor, giving photographers greater flexibility when editing elements such as exposure, color, and white balance.
Apple has its own system-level pipeline for processing RAW files from third-party cameras, exposed to apps through Core Image. It currently includes support and camera-specific calibrations for nearly 800 camera models, with the full and regularly updated compatibility list available here.
Over the years, Apple has updated its RAW processing algorithm eight times, improving how it handles sensor data, demosaicing, denoising, and adjustments such as white balance, exposure, color, and tone.
With iOS 27, Apple is introducing RAW 9, which the company says is “its biggest update yet.” Here’s David Hayward, Core Image Engineer at Apple, on the WWDC26 session Enhance RAW image processing with Core Image:
[RAW 9] dramatically improves the rendering of RAW files. It is built atop a tiled CoreML model, that combines demosaic with denoise for best quality. And the model is run on device using the Apple Neural Engine cores, for optimal performance.
In the session, Hayward shows several examples of RAW 9 in action, comparing the results with RAW 8 and, occasionally, the original, unprocessed sensor data:
This is a zoomed-in crop of a low noise image using RAW 8. This Sony Alpha 7 II image of a vintage dial indicator actually looks quite good. However, when you explore that same image under RAW 9, the image is sharper, clearer, and the fine text is easier to read.
The differences are even more dramatic, when you view high noise images. First, observe the actual RAW data that is contained in this very noisy ISO 51,200 image. In this example from a Canon 5D Mark III, the image is a 10x Crop of a box of crayons. There is so much luma and chroma noise in the RAW data, that it’s impossible to discern the unique color of each crayon. Using our previous algorithms, this is the result! RAW 8 did an acceptable job of recovering the actual colors in the scene. But if you examine the results under RAW 9, the output is significantly better. The colors are accurate and well defined. Even the shiny specular highlights on the crayons are visible.
... continue reading