Skip to content
forked from axodox/unpaint

A simple Windows App for generating AI images with stable diffusion.

License

Notifications You must be signed in to change notification settings

csbarnes/unpaint

 
 

Repository files navigation

Introduction

an image showing a windowed application where the left side shows a painter's workshop with a canvas selected, while the right side shows the same image, but with the selected canvas replaced with a variation of Mona Lisa

This repository contains a fully C++ implementation of Stable Diffusion-based image synthesis tool called Unpaint. It supports txt2img, img2img and inpainting pipelines and the safety checker. This solution does not depend on Python and runs the entire image generation process in a single process, making deployments significantly simpler and smaller, essentially consisting of a few executable and library files (~30MB), and the model weights (~2.5GB / model). Using the library it is possible to integrate Stable Diffusion into almost any application, which can import C++ or C functions.

For technical details please check the page of the C++ Stable Diffusion library which Unpaint is based on here.

System requirements

The application uses DirectML, so it should in theory work with all major GPU vendors, however for now it has only been tested on Windows 11 with an NVidia GPU. We encourage you to try it with different hardware and raise issues on this page if you have problems. Make sure to install the latest Windows updates and especially the latest GPU drivers, as Unpaints benefits from Microsoft's recent Build conference announced optimizations greatly.

Please note that ONNX based pipelines are not yet necessarily as optimized as with using pytorch and CUDA directly, however with latest drivers they are pretty close.

How to install the app?

You can find the app installers on the releases page.

To install the app the first time follow these steps:

  • Download both the .msix and .crt files
  • Right click the .crt file and select install certificate
  • In the wizard select Local Machine and click next
  • Then select Place all certificates in following Store and click Browse
  • Select Trusted Root Certification Authorities and click OK
  • Click Next and then Finish

This installs Unpaint's self-signed security certificate which is required by the new MSIX installer pipeline. As the project grows I will likely purchase a proper paid certificate, but it is around 250USD/yr, and I am doing free work here... Windows Store could be a solution as well, which I am looking into.

If you feel unsafe consider using SignTool to resign the Unpaint package with your own certificate (you need to import your own one as well), you can refer to the end of this file on how to do it.

After this just open the .msix file and should install fine and start the app.

Building the project

Development environment

  • Install Visual Studio 2022
    • Select the following workloads:
      • Desktop development with C++
      • Universal Windows Platform development
      • Game development with C++
    • Also select these individual packages:
      • C++ (v143) Universal Windows Platform tools
  • Install git from here if you do not have it yet
  • I did the project with Windows 11, earlier Windows versions might or might not work

Once you have Visual Studio, for the first build follow these steps:

  • Clone the repo with git clone https://github.com/axodox/unpaint
  • Open Unpaing.sln with Visual Studio
    • If Visual Studio prompts you that you still need to install something, then follow its instructions
    • You could also get a prompt to enable Developer Mode in Windows, which you can do in the Settings app
  • Open the Config folder in the Unpaint project in Solution Explorer, and double click Package.appxmanifest
  • Go to the Packaging tab, and select Choose Certificate
  • Press Create and follow the wizard

After this, you can just start the project and have it be built and run. Enjoy :)

About

A simple Windows App for generating AI images with stable diffusion.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C++ 97.1%
  • PowerShell 1.9%
  • C 1.0%