Skip to content

Latest commit

 

History

History
341 lines (276 loc) · 13.8 KB

[OBSOLETE] krita-sd-local.md

File metadata and controls

341 lines (276 loc) · 13.8 KB

GitHub commit activity GitHub last commit

NOTICE: THIS METHOD IS NOW OBSOLETE AND DEPRECIATED! THERE IS A NEW AND IMPROVED GUIDE.

KRITA + STABLE DIFFUSION PC INSTALLATION GUIDE
FOR WINDOWS 11 & NVIDIA GPUs

Installation of Krita.
Local installation of Stable Diffusion.

Optional Install:

GitHub Desktop

  • You can skip this step entirely and manually download the required zip file later if you prefer. 1
  • I recommend creating a GitHub account, star the repository below (or at least bookmark it in your browser), and install GitHub Desktop:
https://github.com/imperator-maximus/stable-diffusion-webui
  • Install GitHub Desktop:
https://desktop.github.com/
  • Go here:
https://github.com/imperator-maximus/stable-diffusion-webui
  • Click Code > Open With GitHub Desktop.
  • GitHub Desktop default directory should be Documents > GitHub on your PC where it downloaded the folder stable-diffusion-webui.
  • Make note of this directory for later.

TECHNICAL REQUIREMENTS:

Difficulty

Scale:
Easy
Intermediate
[ Advanced ]

There is a Google Colab method that might be added later. I recommend trying that on your own for now as that is an easier, albeit more limiting, method due to Colabs free runtime limitations.

!! - PC/WINDOWS WITH NVIDIA GRAPHICS CARD ONLY !!
!! - Minimum VRAM: 6-8GB
!! - Recommended: 8GB+

  • You need Python, Git, MS Visual Studio, and CUDA for Nvidia GPU all installed.
  • Make sure Windows and all of your Nvidia Drivers are up to date before starting.

Guide Overview

  1. Install Python
  2. (Optional) Install Notepad++
  3. Install Git
  4. Install Microsoft Visual Studio Community 2022
  5. Install Windows Nvidia CUDA Toolkit
  6. Preparing Stable Diffusion Components
  7. Download/Install Krita & Plugin
  8. Krita + SD setup
  9. Getting Started Running SD in Krita
  10. Extras & Acknowledgements

Install Python:

https://www.python.org/downloads/windows/
  • I suggest selecting Copy Path check mark box during installation.

(Optional) Install Notepad++ Editor: 2

https://notepad-plus-plus.org/downloads/
  • Primarily used as an optional editor you can select during the Git installation in the step below.
  • Notepad++ is usually recommended over the default (Vim) - but we won't actively be using it.
  • You can edit .bat (& many more) files with Notepad++ - so it's good to have installed if you ever need to modify those files.
  • If skipped, select Vim during Git install and proceed.

Install Git:

https://git-scm.com/download/win
  • Git Installation Guide:
https://victorkarp.com/git-tutorial-for-beginners-how-to-install-git-on-windows/

Install Microsoft Visual Studio:

https://visualstudio.microsoft.com/
  • Select Download Visual Studio > Community 2022
  • Click on VisualStudioSetup.exe file you downloaded > Select Desktop & Mobile in launcher > select Desktop Development with C+ > Select Install while Downloading > and wait for it to finish. Once completed, you can close any instance of Visual Studio.
  • This is required for CUDA components to run correctly - check your Nvidia GPU card documentation for more information.
  • The primary objective is to get MSVC installed for the CUDA components. This accomplishes that goal.

Install Windows Nvidia CUDA Toolkit:

https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64
  • Information on CUDA installation can be found here:
https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html

Download the Stable Diffusion neural network weights here:

https://huggingface.co/CompVis/stable-diffusion-v-1-4-original
  • The Stable Diffusion model checkpoint file is roughly ~4GB in size.
  • Once downloaded - rename the file from stable-diffusion-v-1-4-original.ckpt to model.ckpt
  • Leave this file in your Downloads folder for now.

(Optional - but recommended)

  • You can use GFPGAN to improve faces, then you'll need to download the model from here:
https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth
  • Leave this file in your Downloads folder for now.

!! DO THIS FOLLOWING STEP ONLY IF:

You didn't Install GitHub Desktop and WebUI folder from earlier.
Otherwise; proceed onward and disregard this.

  • You can skip this step if you did the GitHub Desktop install above. 1
  • Download the Krita Stable Diffusion WebUI Local Interface here:
https://codeload.github.com/imperator-maximus/stable-diffusion-webui/zip/refs/heads/master
  • Extract the stable-diffusion-webui folder from the downloaded zip file into a directory of your choice.
  • Make note of this directory.

Once you have the stable-diffusion-webui folder directory established:

  • Move the model.ckpt and the GFPGANV1.3.pth files to the stable-diffusion-webui directory:
  • If you installed from GitHub Desktop at the beginning, that folder should be in Documents > GitHub > stable-diffusion-webui.
  • If you downloaded the zip file, it's where you extracted the folder to.

Download Krita here:

https://krita.org/en/download/krita-desktop/
  • Select your version > typically Windows Installer 64-Bit > download and install it.
  • Launch Krita after installation, minimize it if needed.

Download and Install Krita Stable Diffusion Plugin here:

https://www.flyingdog.de/sd/en/sd_plugin_krita_beta5.zip
  • Go back to the running instance of Krita.
  • Select Tools > Scripts > Import Python Plugin from File > and select the sd_plugin_krita_beta5.zip file you just downloaded.
  • This will require Krita to restart, so once the plugin is installed, close Krita and relaunch it.
  • The Stable Diffusion panel should be loaded in the bottom right window like so:

image

  • It may take a minute for it to load the new plugin.
  • Once you have confirmed the plugin installed correctly, you can minimize Krita.

!! TROUBLESHOOT: If it did not load correctly !!

  • In Krita, click Settings > Configure Krita > Python Plugin Manager and look for the Stable DIffusion plugin.
  • Make sure the check box on the left is ticked.
  • Click OK and restart Krita - It should load correctly this time.

! If you don't see Stable Diffusion listed - Close Krita > Relaunch > Import the plugin again, and restart Krita.

  • Check the Python Plugin Manager in settings to make sure Stable Diffusion is enabled.


Open the stable-diffusion-webui folder you downloaded earlier:

  • Click on the api.bat file; a windows Command Prompt should launch a local instance of Stable Diffusion on your machine.

!! - If it fails the first time, click on it again to run it.
!! - If it seems to be hanging on any part for long periods I.E. > Installing Requirements for Web UI, select the Command Prompt window and hit Enter to get it going again. If it continues to hang, hit the Enter key again.

  • If everything is successful, you should see in the Command Prompt window output at the end:
> Global Step: 470000
> Model loaded.
> Running on local URL:
http://127.0.0.1:[XXXX]
  • Note down the 127.0.0.1:[XXXX] URL output somewhere (Windows sticky notes or Notion Desktop is nifty).

!! - LEAVE THIS COMMAND PROMPT WINDOW RUNNING TILL YOU ARE FINISHED !!
!! - CLICK api.bat TO RUN YOUR LOCAL SD FROM HERE ON EVERY TIME YOU WANT TO RUN KRITA + SD PLUGIN !!


PUTTING IT ALL TOGETHER AND RUNNING IT:

  • Minimize the CMD Prompt window or move it aside.
  • Open Krita again > on the Stable Diffusion plugin window > click on the Square Icon on the bottom left.
  • A configuration window should appear and under the Webservice URL box enter the URL you noted down earlier.
  • Leave all settings as is and click OK.

FINISHING UP & GETTING STARTED IN KRITA:

  • Click New File on the Krita main page under Start OR hit Ctrl + N OR select File > New
  • Create a new document of any size - 512px X 512px or 1024px X 1024px @ 150-300 Resolution is usually a good place to start.
  • Select the Rectangular Selection Tool in the tool panel on the left - or hit Ctrl + R.
  • Using this tool - draw a box selecting your entire canvas, or size you want the image to be.
  • Select From Text in the Stable Diffusion plugin window.
  • Enter whatever prompt you want to see + modifier words describing the style and detail.
Prompt: An astronaut riding a horse on mars. 
Modifiers: Oil on canvas, detailed, van gough.

kritasdprompt

You can experiment with Steps, Guidance, and Sampling Method for different results:

  • Steps: Higher = Slower but often better quality.
  • Number Images = How many images you want generated per instance - maximum of 4.
  • Advanced - Seed (empty=random) = Random images are generated each time when left blank.
  • Guidance Scale = How strongly the image should follow the prompt.
  • Sampling Method = Various diffusion and sampling methods using the Stable Diffusion model.

Krita_SD

!!!! CONGRATULATIONS !!!! YOU DID IT !!!!

Have fun and experiment with the other options like:

  • From Image = img2img processing where it combines an image + text prompt to generate a new image. You can experiment with how much the image or text influences the overall output by tinkering with settings.
  • Inpaint = Allows you to generate a prompt within an image where there are empty pixels.
    I.E. > You can input an image of a landscape > erase a part of the sky > select it w/ Ctrl + R > Inpaint > and input the text prompt "UFO flying in the sky" and it should give you a selection of outputs to realistically place in that spot.

REMINDER: You need to start api.bat and leave the CMD Prompt window open as long as you are using Krita.
Be sure to close both Krita and the CMD Prompt window when you are done with both.


==== LOCAL WEBUI EXTRAS ====
  • Open stable-diffusion-webui folder where api.bat is.
  • Look for a file named webui.py
  • Open with a basic text editor like Notepad (or Notepad++ if you decided to install that earlier ^_^) 2
  • Scroll towards the end (near line 90) and look for:
demo.launch(

And change to:

demo.launch(share=True)(
  • Save and exit > in stable-diffusion-webui folder> Click webui.bat.
  • The familiar CMD Prompt window should pop up and launch webui.bat just like it does api.bat.
  • At the end, the output should be:
> Model loaded.
> Running on local URL: http://127.0.0.1:[XXXX]
> Running on public URL: https://[RANDOM_NUMBER].gradio.app
  • Open up your favorite browser (Firefox, Ungoogled Chromium, or Brave are good.)
  • Copy and paste the gradio.app web address and go.
  • The browser should load a local webui instance of your very own Stable Diffusion WebUI!
  • The CMD Prompt window functions just like api.bat.

Image of gradio.app UI in browser image

  • This panel functions similar to the Krita plugin; txt2img, img2img, and tons of diffusion options.
  • There are far more technical options under Settings.

Image of gradio.app UI on mobile

310648807_823000495368859_1392061855349642323_n

  • And it even works outside your own network; Just enter the XXXX.gradio.app address in any browser!
  • Close the CMD prompt webui.bat window when you are done running this instance of Stable Diffusion.


==== ACKNOWLEDGMENTS ====

This guide was assembled from the following websites:


DONATIONS SO I CAN KEEP ON TINKERING

If you found this useful:

  • ETH:
0x97Fb39F4d9e1b1aE6ad9D3Cf99fb7C423Cb5FafE
  • BTC:
bc1q073hyc4gnd4zpr3zvxldqxd7pwusktw7tguu4g
  • SOL:
4CeqibD5LqpPb5XgC9TdxfA2NMUBntnUTRFTEWzvBzVo
  • XMR:
88hWjDuptnBerfkoTTAUhJ4AFuiMnMPSVVQhAbiV2rSEV7Gj2FaytRv1bnL8gPmL6U4L4XhFVBc4KbQLDmDM9hEaC4S1FV5

==== EVERY BIT HELPS! ====


==== THANKS FOR READING! ====