AI Briefing
KO

SD Support for Apple Silicon

·2022.12.01 09:00

Key point

Reveals how to run Stable Diffusion on Apple Silicon via Core ML and releases the converted models.

Details

In collaboration with Apple engineers, released models and scripts converted to Core ML to run Stable Diffusion on Apple Silicon.

Converted checkpoints for Stable Diffusion v1.4, v1.5, v2 base, v2.1 base are provided via the Hugging Face Hub, and Core ML allows use of the CPU, GPU, and Neural Engine (NE) all together.

Users can choose conversion options that deliver optimal performance depending on their hardware:

  • Attention type: split_einsum, which is compatible with all units, and original, which is CPU/GPU only.
  • Model variant: packages for Python inference and compiled versions for Swift code are provided separately.

It delivers optimal performance on macOS Ventura 13.1 or later, and image generation takes about 18 seconds on an M1 Max.

computer-visionhuggingfaceapple-siliconcore-mlstable-diffusion
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

GeekNewsUnderdog Releases 'Husky', a Model-Specific Inference Engine Up to 4.5x Faster than MLXTLDR AIJared Palmer Releases 'Kev', a Lightweight Decision-Making Model Based on Qwen3.5HuggingFace BlogHugging Face Adds GGUF Model Support to TransformersHuggingFace BlogHugging Face Hires oMLX Developer to Expand MLX SupportPyTorchKRvLLM Releases 'Decision 1.0', AI Models Specialized for Decision-MakingRedditHemmingway-1, a 27B Creative Writing Model Based on Qwen3.8, Released