/dreambooth-gui

Primary LanguageTypeScriptMIT LicenseMIT

Dreambooth Gui

Provides a easy-to-use gui for users to train Dreambooth with custom images. This Gui supports any NVIDIA card with >10GB VRAM.

BuildStatus

Screenshots

203261990-a1920337-1687-4503-84bd-ffcd599d3c42 203261994-f541593a-65e1-4a4e-9acc-6a05d5ce310d 203261997-cc3c8456-a7fc-4162-b4fd-d953b4c5fbcd

Highlights

  1. Automatically decide training params that fit your available VRAM.
  2. Easy to use Gui for you to select images.
  3. Support prior preservation training with class images.
  4. Automatically cache models.

Install (Windows)

  1. Download and install docker from https://www.docker.com/
  2. Setup WSL2 for windows. https://learn.microsoft.com/en-us/windows/wsl/install
  3. If you find 'WSL 2 installation is incomplete' when starting docker, you can follow this video to fix it. https://www.youtube.com/watch?v=Ffzud5xLH4c
  4. Download and install dreambooth-gui_*_x64_en-US.msi from release page.
  5. Run the dreambooth-gui as administrator.

Install (Linux)

  1. Download and install docker from https://www.docker.com/
  2. Download AppImage from release page.
  3. Run chmod +x dreambooth-gui_*amd64.AppImage
  4. Run sudo dreambooth-gui_*amd64.AppImage

FAQs

  1. Failed to create directory

    Please make sure you have the latest verion of GUI. This is a old bug that fixed in v0.1.3

  2. PIL.UnidentifiedImageEnnon: cannot identify image file

    Make sure the instance image folder only have image.

  3. Read-only file system error

    Make sure you have enough space in C(or home folder) before running the Gui.

  4. Train with SD v2

    Training with SD v2 is supported. However, you need to type stabilityai/stable-diffusion-2 as model name. Local v2 training is not supported right now.

  5. I have other questions!

    Please use the discussion page for Q&A.

    I will convert FAQs to a bug if necessary. I perfer to keep the issue section clean but keep getting questions that I answered before.

Roadmap

  • Refactor the state management.
  • Better error handling to cover FAQs.
  • Allow advanced customization
    • Load local model.
    • Save/Load config for users.
    • Save model / pics in places other than $APP_DIR
  • Better training progress report.
    • Create a dialog when training finished.
    • Progress bar.
  • Support model converstion.

Additional Resources

  1. Someone in japan write a doc regarding how to use it.