Apple has announced a major update to its system-level RAW image processing engine with the release of iOS 27, iPadOS 27, and macOS 27. Dubbed RAW 9, this new version leverages machine learning to dramatically improve detail and reduce noise, even when reprocessing older RAW photos. The company claims it is the biggest update to its RAW processing algorithm yet, and early examples demonstrate a significant leap in image quality that will benefit both professional photographers and enthusiasts.

Understanding RAW Photography

RAW is an image format that preserves all data captured by a camera's sensor, without applying in-camera processing like white balance, sharpening, or noise reduction. This gives photographers maximum flexibility during post-processing, allowing them to adjust exposure, color temperature, and other parameters without degrading image quality. However, RAW files require sophisticated demosaicing and denoising algorithms to produce a viewable image. Demosaicing reconstructs full color information from the Bayer pattern or other sensor filter arrays, while denoising removes the random fluctuations inherent in electronic sensors. Apple's Core Image framework provides a system-level pipeline for processing RAW files from third-party cameras, supporting nearly 800 camera models with specific calibrations tailored to each sensor's unique characteristics.

The Evolution of Apple's RAW Engine

Since introducing its RAW processing pipeline, Apple has updated it eight times, each iteration improving how sensor data is handled. RAW 1 through RAW 8 progressively enhanced demosaicing, denoising, white balance, exposure, color, and tone mapping. With each update, Apple refined the algorithms to better handle the unique characteristics of different camera sensors, such as the varying color depth, dynamic range, and noise patterns. The introduction of RAW 9 marks a paradigm shift, as it moves away from traditional signal-processing algorithms to a machine learning-based approach. This is not just an incremental improvement; it represents a fundamental change in how Apple processes RAW data, leveraging the power of the Apple Neural Engine to deliver results that were previously unattainable with conventional methods.

What RAW 9 Brings to the Table

According to David Hayward, Core Image Engineer at Apple, RAW 9 is built atop a tiled CoreML model that combines demosaicing with denoising for the best possible quality. The model runs on-device using the Apple Neural Engine (ANE) cores, ensuring optimal performance without sacrificing battery life. This allows for real-time processing of RAW images within apps that use Core Image. The tiled approach means the image is processed in overlapping patches, which enables handling of very large files while maintaining memory efficiency. By integrating demosaicing and denoising into a single neural network, RAW 9 can make better decisions about which pixels to smooth and which to preserve, resulting in sharper details and fewer artifacts.

The key improvement is in handling high-noise images. At high ISO settings, sensor data is often overwhelmed by luma and chroma noise, making it difficult to recover accurate colors and details. Traditional algorithms struggle to distinguish noise from fine texture, often blurring out important details. RAW 9's machine learning model is trained on vast datasets to recognize patterns and reconstruct the underlying image with significantly less noise and more accurate color reproduction. This is especially beneficial for photographers who shoot in low light or use older cameras with noisier sensors.

Real-World Examples That Show the Difference

Apple showcased three compelling examples during the WWDC26 session on Core Image enhancements. First, a low-noise image captured with a Sony Alpha 7 II of a vintage dial indicator. Under RAW 8, the image already looked good, but RAW 9 produced a noticeably sharper and clearer result, making fine text easier to read. This demonstrates that even in optimal conditions, the new algorithm extracts more detail.

Second, a high-noise image from a Canon 5D Mark III at ISO 51,200. The RAW data was almost unusable, with so much luma and chroma noise that the colors of individual crayons in a box were indistinguishable. RAW 8 did an acceptable job of recovering the scene, but RAW 9 delivered a stunning improvement: accurate, well-defined colors and even shiny specular highlights visible on the crayons. This example highlights the power of machine learning to recover information that traditional algorithms would lose.

Third, an image taken with a Fujifilm X-T5 at ISO 12,800. This camera uses a non-traditional X-Trans sensor pattern, which is notoriously challenging for demosaicing. Under RAW 8, the result showed color artifacts and loss of detail in embroidery yarn. RAW 9 eliminated these artifacts, producing a cleaner image with more legible small text and clearer texture in the yarn. This shows that RAW 9 is not just better for conventional Bayer sensors but also excels with exotic filter arrays.

Implications for Photographers and Developers

For photographers, RAW 9 means they can expect better results from their existing RAW libraries, even from cameras they no longer own. The reprocessing capability is a major boon, as users can revisit old files and extract more detail and less noise than ever before. The update is available across iOS 27, iPadOS 27, and macOS 27, ensuring consistency across Apple devices.

For developers, Apple provides a straightforward API to enable RAW 9 in their apps. The WWDC session goes into detail on how to integrate the new engine, optimize performance for editing workflows, and handle exporting. Developers are encouraged to adopt the new processing mode to give their users the best possible image quality. The session also covers tips for managing memory and using the Neural Engine efficiently, ensuring that even complex edits remain responsive.

Broader Context: iOS 27 and the Machine Learning Ecosystem

The introduction of RAW 9 is part of a broader trend in iOS 27 that emphasizes on-device machine learning. Apple has been investing heavily in the Neural Engine and CoreML, enabling new features across the system. RAW 9 is just one example; others include enhanced photo editing tools, improved computational photography features like Smart HDR and Deep Fusion, and more intelligent camera scene detection. Apple's approach keeps processing local, preserving user privacy and reducing latency compared to cloud-based alternatives. This integration of machine learning into core system capabilities is a strategic advantage, allowing Apple to deliver professional-grade tools without requiring specialized hardware.

As Apple continues to push the boundaries of computational imaging, RAW 9 represents a significant step forward. The combination of traditional signal processing with machine learning promises to deliver the highest quality RAW conversions yet, making it an essential update for any photographer using Apple devices. The examples from the WWDC session leave little doubt that this is a transformative improvement that will change how users interact with their RAW files.

Source: 9to5Mac News