Mochi Diffusion Homepage, Documentation and Downloads – Running Stable Diffusion Natively on Mac – News Delivery

Mochi Diffusion runs Stable Diffusion natively on Mac,This app has built-in Apple’s Core ML Stable Diffusion Framework ,In order to achieve the best performance with an extremely low memory footprint on Macs equipped with Apple chips.

Function

  • Extreme performance and extremely low memory footprint (~150MB when using the neural network engine)
  • Take advantage of the Neural Engine on all Macs with Apple silicon
  • Generate images without internet connection
  • Image to Image (also known as Image2Image)
  • Store all keywords in the image’s EXIF ​​info (view in Finder’s “Show Info” window)
  • Scale up the resulting image using RealESRGAN
  • Automatically save & restore images
  • Custom Stable Diffusion Core ML Models
  • No need to worry about damaged models
  • Developed using macOS native framework SwiftUI

download

exist issued page to download the latest version.

run

When running the model for the first time, the neural network engine may take about 2 minutes to compile the cache, and the subsequent running speed will be significantly improved.

computing unit

  • CPU 和神经网络引擎 Good balance between performance and memory usage
  • CPU 和 GPU Might be faster on M1 Max/Ultra and later models, but use more memory

You need to select the corresponding model according to different computing units (see the model section for details).

Model

You need to convert or download the Core ML model yourself to use Mochi Diffusion.

Here are a few converted models uploaded

  1. convert or download the Core ML model

    • split_einsum Versions are available for all Compute Units including the Neural Engine
    • original version only works with CPU 和 GPU

  2. By default, the application’s models folder will be created in your home directory.This location can be customized under Settings
  3. In the model folder, you can create a new folder, rename it with the name you want to display in the app, and then put the converted model into the folder
  4. Your folder path should look like this:

<主目录>/
└── MochiDiffusion/
    └── models/
        ├── stable-diffusion-2-1_split-einsum_compiled/
        │   ├── merges.txt
        │   ├── TextEncoder.mlmodelc
        │   ├── Unet.mlmodelc
        │   ├── VAEDecoder.mlmodelc
        │   ├── VAEEncoder.mlmodelc
        │   └── vocab.json
        ├── ...
        └── ...

compatibility

  • Apple Silicon Macs (M1 and later)
  • macOS Ventura 13.1 and above
  • Xcode 14.2 (self build)

privacy

All calculations are done locally and absolutely no data is uploaded.

#Mochi #Diffusion #Homepage #Documentation #Downloads #Running #Stable #Diffusion #Natively #Mac #News Delivery

Leave a Reply

Your email address will not be published. Required fields are marked *