Apple’s ninth RAW processing engine, shipping with OS 27 software, integrates machine-learning denoising at the demosaicing stage for sharper results across Apple hardware and hundreds of supported cameras.
As is often the way with its WWDC event, some of Apple’s more technically interesting announcements about its forthcoming software take place away from the main stage. There are, after all, over 250 software changes coming across its various platforms, and there’s only enough time to sketch out the key ones.
But the video titled Enhance RAW image processing with Core Image published from its Apple developer account is particularly interesting, as it walks through RAW 9, the ninth iteration of Apple’s Core Image RAW processing pipeline. And it looks like the company isn’t underselling it when it calls this its “biggest update yet.”
Machine Learning denoising
As surfaced by DPReview, the centerpiece is machine-learning-powered denoising integrated directly into the demosaicing step, rather than being applied afterwards as a separate pass. Apple says this produces sharper results across the board, with dramatic improvements when dealing with very noisy files.
Given the leg up that will give the iPhone’s relatively small sensors, that’s a significant advance. But RAW 9 isn’t limited to Apple hardware: the company says it will support RAW files from currently 784 cameras across all major professional vendors (the demos feature the Sony Alpha 7 II, Canon 5D Mark III and Fujifilm X-T5), with more to be added over time.
There’s a clear difference in quality between this RAW 8 image and the RAW 9 one at the top of the article
AI-assisted denoising is becoming increasingly ubiquitous. Photoshop, DxO, and specialist tools like Topaz Labs have all leaned into it heavily. So, Apple baking it into the core pipeline is a logical step, though notably in its current iteration it offers limited user-facing controls. Apps will be able to expose a luma noise reduction slider, but little else. Also, the color noise reduction slider has been jettisoned as the CoreML model handles chroma noise automatically.
Deployment this fall
RAW 9 ships with iOS 27, iPadOS 27, visionOS 27, and macOS 27 “Golden Gate” this fall and Apple’s own apps will use it automatically. The new engine will be able to be applied retrospectively to existing archives too, which should keep anyone with large libraries happily engaged for a while when the software releases.
Third-party developers can opt in, though it won’t be the default, likely reflecting the higher processing demands compared to RAW 8. Adobe and other vendors with already developed custom RAW engines are unlikely to adopt it. But the ever-growing band of indie image editing apps on Apple platforms stand to gain meaningfully improved output with minimal extra development work.
Leave a comment