Mochi Diffusion
A local AI image generator for macOS that runs Stable Diffusion and FLUX.2 Klein offline on Apple Silicon. Open source under GPL-3.0.
Mochi Diffusion is a native macOS image generator for people who want to create images on an Apple Silicon Mac without sending prompts or results to a cloud service. It runs Stable Diffusion and FLUX.2 Klein locally, with completely offline generation. The app collects no telemetry or other data and is open source under GPL-3.0.
Its Stable Diffusion support uses Apple's Core ML implementation, which can use the Neural Engine to keep memory use low. CPU and GPU processing is also available and may be faster on some Macs, though it uses more memory. Models need to be converted to Core ML format, and each converted model has a fixed output size. Community models and ControlNet models are supported, along with generation based on a starting image.
FLUX.2 Klein runs through the Iris pipeline with Metal GPU acceleration and doesn't require model conversion. The app supports the distilled 4B and 9B models. Klein can use multiple input images in a generation; the app automatically resizes inputs when necessary and shows their adjusted dimensions.
The SwiftUI interface includes a gallery with image import, saving and sync support. Saved images retain their prompts in EXIF metadata, so the prompt stays attached to the image file and can be inspected in Finder. The app requires an Apple Silicon Mac, starting with M1.
