SDXL Core ML Support for Apple Silicon

Apple and Hugging Face have released Core ML porting and optimization techniques to support running Stable Diffusion XL (SDXL) smoothly on Apple Silicon Macs.

The key updates are as follows:

• Core ML Porting: The SDXL base model has been ported to Core ML, enabling direct use in native Swift-based apps.
• Mixed-bit Palettization: Advanced compression techniques reduced the UNet size from 4.8GB to 1.4GB, an approximately 71% reduction, while minimizing quality degradation.
• Tool and Demo Updates: Both Apple's model conversion/inference repository and Hugging Face's Swift demo app have been updated, allowing users to convert or test models directly.

Users can immediately use already-optimized Core ML models by downloading them from the Hugging Face Hub, and can also convert their own fine-tuned models to Core ML using the updated tools.