/DiffusionToolkit

Stable Diffusion Image and prompt database

Primary LanguageC#MIT LicenseMIT

Diffusion Toolkit

Diffusion Toolkit is an image viewer built in .NET that scans your images for PNGInfo generated by diffusion image generators like AUTOMATIC1111, NovelAI, NKMD and others.

With Diffusion Toolkit you can:

  • Select and recursively scan a set of folders for PNG and JPG+TXT images.
  • Parse the metadata (prompt & parameters) into a database.
  • Search based on prompts and parameters, even file creation date (see Advanced Searching)
  • Quickly and easily mark files for later deletion
  • Tag images as Favorite
  • Drag and drop from Diffusion Toolkit to A1111 or NovelAI PNGInfo, or any app for folder to copy the image to the drop target.

Diffusion Toolkit supports A1111 and NovelAI PNGInfo formats with the following parameters:

  • prompt
  • negative prompt (uc)
  • steps
  • sampler
  • seed
  • scale (CFG scale)
  • model hash
  • aesthetic_score
  • hypernet & hypernet_strength

Diffusion Toolkit also tracks:

  • Created date
  • Path

Think Lexica.art, but with more powerful search on your local images.

Themes

Diffusion Toolkit supports Light and Dark themes. By default, it will use your system theme.

image

image

Installation

If you want to just run the program, get the latest release from here. Look for the Assets under the latest release, expand it, then grab the zip file.

This is Windows only. You may be required to install .NET 6 Desktop Runtime (https://dotnet.microsoft.com/en-us/download/dotnet/6.0)

TODO

  • Snippets and prompt management
  • Manage Checkpoint metadata alongside file
  • Advanced Search UI?

Future

  • Embed Checkpoint metadata inside checkpoint for use in SD UIs

Getting Started

See the in-app documentation for Getting Started

Troubleshooting

If you get some sort of database error, try deleting the file diffusion-toolkit.db in %APPDATA%\DiffusionToolkit

Buy me a coffee

I've been told that people should be allowed to buy me a coffee

Beer works too (PayPal)