Apple, once skeptical of generative AI photo editing over reality-distortion concerns, unveiled a suite of AI image manipulation tools at WWDC 2026. The move marks a fundamental strategic shift, putting Apple on par with Google Photos and Samsung, which have offered similar features for years. The new tools—expected in iOS 27—will give users effortless image manipulation capabilities, reigniting debates around deepfakes and photo authenticity.
Apple announced CoreAI at WWDC, which the post frames as a possible future replacement for CoreML and an alternative to MLX, llama.cpp, and torch for optimized on-device inference. Models still need conversion through Python scripts, and current supported models appear mostly from mid-2025. No performance data is available yet; the author expects it may trail MLX on GPU, but Apple’s 20B on-device foundation model claim suggests larger app-bundled models could become possible.
Apple’s Core AI framework is positioned as a developer stack for deploying AI models directly inside apps on Apple silicon. The documentation describes Swift APIs, `.aimodel` assets, model specialization, caching, Xcode profiling, and debugging tools. It appears aimed at developers building low-latency, privacy-conscious on-device inference workflows, though the documentation is marked as preliminary beta information.