Mochi Diffusion is a native macOS application that runs Stable Diffusion and FLUX.2 Klein image generation models on Apple Silicon Macs.
The tool solves the problem of running computationally intensive diffusion models on consumer Mac hardware by leveraging Apple's Core ML framework and the Neural Engine. It converts standard Stable Diffusion models to Core ML format, which allows the Neural Engine to execute inference efficiently while keeping memory usage low. The application generates images locally and completely offline, supporting text-to-image generation, image-to-image transformation, and ControlNet-based image guidance. Generated images are automatically tagged with prompt metadata in EXIF format for easy reference.
Mochi Diffusion suits developers and creators working on Apple Silicon Macs who want fast, private image generation without relying on cloud services or external APIs. The tool is particularly valuable for those with M1, M2, M3, or later chips, as it fully utilizes the Neural Engine to achieve speeds around 150MB memory usage. Users should be aware that first-time model compilation can take up to two minutes as the Neural Engine caches an optimized version, but subsequent generations run much faster. The application offers flexibility in compute unit selection, allowing users to choose between CPU and Neural Engine for balanced performance and low memory, or CPU and GPU for potentially faster generation on higher-end chips like M1 Max and Ultra, though at the cost of increased memory consumption. Custom Core ML models can be imported, and the built-in gallery supports importing, saving, and syncing generated images.
Development activity shows consistent engagement with the project through community contributions and localization efforts across multiple languages. The maintainers actively curate community model resources and provide documentation for converting and using different model variants. The project maintains active communication channels and accepts community feedback for feature development and model support.
