WWDC 26: Apple Intelligence Revolutionizes Image Playground with Photorealistic Images and Private Cloud
On stage at WWDC, Apple presented a completely revamped version of Image Playground, powered by new generative models running on private cloud compute. The most significant update is support for photorealism: until now, Image Playground was tied to illustrative styles, but it is now possible to generate images in virtually any style, including photorealistic ones.
The workflow described on stage allows users to start from a person in their photo library, add a natural language text description, and obtain a composite image. From there, it's possible to refine the result with additional text instructions or direct gestures on the screen: circling an object selects it, allowing movement, resizing, or modification through text. In the demo, a birthday cake was shown with candles being added in this manner.
Regarding privacy, Apple made it clear that the photos used in the process are never stored or shared, not even with Apple itself. Image Playground is system-integrated and will be available in new contexts compared to the previous version: generating contact posters, lock screen wallpapers, and message backgrounds. The system also offers personalized suggestions based on the user's photos, referencing recurring places and activities in the roll. The new features are also exposed through the Image Playground API for developers.
Photos: three new editing tools with Apple Intelligence Alongside Image Playground, Apple announced three new editing tools in the Photos app, all based on the new Apple Intelligence models and private cloud compute. The first is an update to the existing Cleanup tool, with improved quality of disturbance element removal and more realistic filling even in complex scenes.
The second tool is Extend, which allows users to expand the edges of an image to give more breathing room to the main subject or to correct a crooked horizon without sacrificing portions of the original frame.
The third tool, described as the most sophisticated, is Spatial Reframing. This tool combines on-device spatial models, developed also through experience with Apple Vision Pro, with generative models on private cloud compute. The result is the possibility to virtually move the viewpoint after the photo has been taken: by dragging the image in the new editing interface, the perspective adjusts as if the camera were physically repositioned in the original scene. Areas that fall outside the original frame are filled by generative models, generating only the content necessary to fill the gaps created by the shift. Apple clarified that the feature is compatible with almost any photo in the library, including older images or those taken with cameras from other manufacturers.