High dynamic range (HDR) photography techniques can accurately capture a scene’s full range of brightness values — from those of its darkest shadows to its brightest light sources. However, when the time comes to view the resulting HDR photos, many displays are only capable of showing a limited range of brightness levels. This discrepancy means that HDR photos are often downgraded to fit within the physical limitations of these standard dynamic range (SDR) displays and the corresponding file formats that store a limited range of values.
However, an increasing share of mobile device and computer displays are HDR-capable, showing a greater range of brightness levels with brighter whites and darker blacks. Although several common HDR file formats capable of encoding this greater range exist, HDR-only formats can appear differently depending on the display capabilities. To optimally show images on any display — no matter the dynamic range — Android 14 introduced the new Ultra HDR image format in October 2023. Google Pixel 7 and newer devices adopted the format, capturing and saving HDR photographs as Ultra HDR and affording a more true-to-life rendition as compared to SDR formats, with greater contrast and brighter highlights.
Although the format unlocked HDR image display, its new metadata also increased the complexity of image editing operations. Until now, if you applied a more complex edit like Magic Eraser to an Ultra HDR photograph using Google Photos, your new edited image would save in SDR, losing its HDR rendition and the associated brightness and contrast.
Today, we introduce a new machine learning (ML) technique that enables complex image editing for HDR photographs, including those saved as Ultra HDR. On Pixel 8 and newer, our model runs behind the scenes in Google Photos to ensure that even if you apply complex edits to HDR images, your new edited images remain HDR. Core to our technique is predicting the HDR image metadata missing after editing, using an ML model trained on a large dataset of HDR images with complete metadata.
