Hugging Face has released version 1.1 of its Diffusers for Mac app, a native image generation tool that leverages Core ML optimization to deliver up to twice the performance of previous versions. The app, available on the Mac App Store and open source on GitHub, allows users to generate images from text prompts entirely offline, using state-of-the-art diffusion models from the Hugging Face Hub.
The key improvement in version 1.1 is automatic accelerator selection. The app now intelligently chooses between the GPU and Apple Neural Engine (ANE) based on the user's Mac hardware. Benchmarks show that on standard M1 Macs, the ANE delivers roughly double the speed of the GPU, while on high-end M1 Max machines with 32 GPU cores, the GPU is significantly faster. The app also supports simultaneous use of CPU, GPU, and ANE for maximum throughput.
Additional features include guidance scale adjustment, optional safety checker disable, generation cancellation, model download indicators, and a seed reuse shortcut for tweaking parameters. The app remains fully local and private, with no data leaving the user's computer.
Hugging Face is calling on the community to submit benchmark data from M1 Pro, M2 Pro, and M2 Max devices to further refine performance heuristics. Future updates are planned to bring easier access to fine-tuned models from the Hub and iOS/iPadOS support.