MochiDiffusion enables native Stable Diffusion image generation on macOS, specifically targeting Apple Silicon Macs. It provides a fast, memory-efficient, and offline solution for users looking to leverage their Mac hardware for AI art creation, including features like Image2Image and ControlNet.

How It Works

MochiDiffusion utilizes Apple's Core ML framework, optimizing Stable Diffusion models for Apple Silicon's Neural Engine. This approach maximizes performance and minimizes memory usage (around 150MB with the Neural Engine), offering a significant advantage over CPU-bound or less optimized GPU solutions on Mac hardware. The use of compiled Core ML models ensures faster inference after an initial compilation cache.

Quick Start & Requirements

• Install: Download the latest macOS native app from the project's releases.
• Prerequisites: macOS Sonoma 14.0 or later. Xcode 15.2 is required for building from source.
• Models: Requires Core ML converted Stable Diffusion models. Instructions and community models are linked.
• Setup: Initial model compilation may take up to 2 minutes per model; subsequent generations are faster.

Highlighted Details

• Native macOS app built with SwiftUI.
• Extremely memory efficient (~150MB VRAM with Neural Engine).
• Supports Image2Image and ControlNet.
• Generated images include prompt info in EXIF metadata.
• Upscaling capabilities using RealESRGAN.

Maintenance & Community

The project welcomes contributions via bug reports, code, and translations. Issues and feature requests can be submitted via GitHub Issues. Translations are managed on Crowdin.

Licensing & Compatibility

The project's licensing is not explicitly stated in the README. Compatibility is limited to Apple Silicon Macs running macOS Sonoma 14.0+.

Limitations & Caveats

The README does not specify the license, which may impact commercial use or integration into closed-source projects. Initial model compilation can be time-consuming. The project relies on Apple's Core ML implementation, which might have its own performance characteristics or limitations compared to other platforms.