Core ML optimizations enable faster Stable Diffusion on iPhone, iPad, and Mac

AI Impact Summary

Apple's Core ML updates enable on-device Stable Diffusion inference with 6-bit palettization, reducing model size and memory footprint while preserving speed. The runtime uncompresses weights on-the-fly during inference and leverages optimizations like SPLIT_EINSUM_V2 to boost Neural Engine performance. Developers can convert PyTorch/TensorFlow models with coremltools.optimize and ml-stable-diffusion, targeting iOS/iPadOS 17 or macOS 14 to access these capabilities; Apple has pushed four 6-bit Stable Diffusion variants to the Hub to accelerate adoption.