SD 1.5 CoreML conversion for Apple Silicon
1 min readupdated
An unofficial version of the classic Stable Diffusion v1.5 image generation model optimized for Apple Silicon processors. The architecture uses mixed Float16 precision and 6-bit palletization (a weight compression method to reduce memory consumption), resulting in a model size of around 1.5 GB. This solution is designed to run generations directly on Mac devices without discrete graphics cards.
Key features
- Optimized for Apple Silicon using Core ML
- Compact size of about 1.5 GB thanks to 6-bit palletization
- Includes all pipeline components: UNet, CLIP Text Encoder, VAE, and Safety Checker
- Supports the standard 512 by 512 pixel resolution, negative prompts, and the CFG parameter
What you can do
- Run local image generation on a Mac computer via the AIATOR app
- Explore how quantized Core ML models work within the Apple ecosystem
Example
To use it, download the repository files to the ~/Library/Application Support/AIATOR/models/coreml-sd15 folder and select SD 1.5 in the AIATOR studio settings.
Read more
DetailsBase model: runwayml/stable-diffusion-v1-5
Conversion tool: Apple Core ML Stable Diffusion tooling
Size: ~1.5 GB
License: CreativeML OpenRAIL-M










Comments