Apple releases CoreML-Next with native diffusion acceleration for MacBooks

Tech · 4 min read

Apple unveiled CoreML-Next, an update to its machine learning runtime designed to run large diffusion and multimodal models efficiently on Apple Silicon. Improvements include optimized memory tiling, mixed-precision pipelines, and energy-aware schedulers that keep creative workflows responsive without thermal throttling.

For designers, this means AI features like in-app image generation, on-device upscaling, and low-latency style transfer can run directly within native apps such as Sketch-style editors and compositors. Apple also added new APIs for layer-aware inference so models can operate on vector layers and raster layers distinctly.

Apple is distributing CoreML-Next in Xcode today and is inviting third-party devs to submit apps that leverage the new capabilities to the App Store. The company framed the release as a privacy-forward approach for designers who prefer local inference over cloud-based services.