How to install and run Stable Diffusion on your Windows PC
Background details are at the end of this article.
HOW TO INSTRUCTIONS: 8 STEPS
This article assumes you want to install and run your own Stable Diffusion AI Art Generator and that you are or will be using the following hardware / software versions as of November 7th 2022:
- PC with Windows 10 or newer
- NVIDIA RTX + or GTX 1660ti+
- NVIDIA Drivers are up to date
- Intel Core i7 or newer
- RAM: 16gb+
- 64 bit operating system
- Git Bash
- GitHub Account
- Windows PowerShell
- Python 3.10 (note: 3.11 will NOT work for Tensorflow in this setup.)
- Pip 22.3.1 +
- RUN GIT BASH from inside your local github directory.
2. GIT CLONE this repo to your local computer’s github directory: https://github.com/monigarr/mohawklanguage
3. GIT BASH: Check if you have Python installed and note what version of python you have installed. Use python 3.10 or older. Python 3.11 is NOT working with Tensorflow as of 11/7/2022
- $command_line: python — version
If you do not have python installed; install python and note what version you installed. Note: you want python 3.10 64bit or older. Do NOT use python 3.11+
4. INSTALL PIP: move to the folder with ‘get-pip.py’ file and install pip.
- $command_line: cd mohawklanguage/MoniGarr_AI_Experiments/Diffusion/MoniGarr_Tutorials/Stable_Diffusion_MoniGarr_V0.1
- $command_line: python get-pip.py
5. RUN POWERSHELL as administrator and enter the following command to enable long file path support:
- $command_line: New-ItemProperty -Path “HKLM:\SYSTEM\CurrentControlSet\Control\FileSystem” ` -Name “LongPathsEnabled” -Value 1 -PropertyType DWORD -Force
6. INSTALL DEPENDENCIES, by entering the following in your command line window:
- $command_line: pip install keras-cv
- $command_line: pip install tensorflow
- $command_line: pip install tensorflow_datasets
- $command_line: pip install matplotlib
- keras
- tensorflow
- datasets
- matplotlib
7. MOVE TO THE FOLDER that holds ‘ stable_diffusion_monigarr_v0.1.py ‘ file, if you are not allready their. ( D:\GITHUB_PROJECTS\mohawklanguage\MoniGarr_AI_Experiments\Diffusion\MoniGarr_Tutorials\Stable_Diffusion_MoniGarr_V0.1
$command_line: cd …./Stable_Diffusion_MoniGarr_V0.1
8. RUN python stable_diffusion_monigarr_v0.1.py
- $command_line: python stable_diffusion_monigarr_v0.1.py
Some Stable Diffusion Art I generated with the above instructions on my own windows pc. Note: I did NOT input any of my own original art to generate these two images with the following text prompt that is hard coded into the python file: “ Forest landscape with distant river”
Some Art I created with my own digital art inputs and Disco Diffusion:
BACKGROUND:
Stable Diffusion AI Art Generators are currently found in many different versions all over the internet and run from a variety of discord servers, web browsers, jupyter notebooks, google colabs, huggingface api keys and public datasets. Stable Diffusion is also available as extensions and add-ons for Krita, Blender, UnReal Engine and more. Artists and non-Artists use AI Art Generators to quickly produce brand new digital art that never existed before. Most AI Art Generators are trained on public datasets that are created with public web scrapers that take human produced art (visual, audio); without prior informed consent from the humans they scrape from. The massive amounts of human produced art that are scraped up is then used to create datasets that are then used to train data models that make it possible for AI Art Generators to produce brand new art that has never existed before. The exploitation of human artists without their prior fully informed consent is shaking up every industry with questions about legalities, morals and the consequences (or lack of consequences) for those that architect tech solutions designed to ignore legalities, moral issues while providing financial & non-financial benefits to those that build and host AI Art Generators.
I share this information with the sincere hope of lessening and perhaps stopping the fast growing digital divide that pushes out specific people / communities, entire cultures and demographics from being included, from leading and from participating in these conversations, actions and projects. I know all AI Industries can benefit from healthy emotional intelligence tactics of acknowledging & respecting those that are currently & potentially harmed and exploited because of the willful act of removing full prior informed consent.
Exploitation in this context also refers to the lesser known acts of
- web scraping
- literal complete erasure of human artists existence
- replacement of human artists names & works by AI generated content.
The exploitation description is not an exaggeration; human artists names & works are literally being erased & replaced by AI Generators as of November 2022.
What can you and I do to correct the harmful situation that AI Art Generation is creating? My questions are heavily influenced by Kanien’kehake, Onkwehonwe, Haudenosaunee relatives and friends. Here are a few bits of what I often ask myself working in tech industries.
SELF QUESTIONING 1: Why can’t I build and maintain art datasets that are designed with respect, prior fully informed consent and fair compensation of each & every artist who’s work and names are in the dataset?
Instead of secret web scraping, erasing and replacing of every artist in existence for my own financial & non-financial benefits — — why not produce respectful, unique, niche datasets with many human beings who are all sincerely joyful, reciprocated and respected? This is what I call Onkwehonwehneha A.I. A very rough / loose translation from Kanien’keha of my practice is a ‘Native way of being / doing; where Ancient Intelligence is teaching and guiding Artificial Intelligence to bring the best it has to the table’. Rather than teaching and guiding Artificial Intelligence to secretly web scrape, erase & replace human beings works with disrespect.
SELF QUESTIONING 2: Just because I can, Should I?
As a software dev and artist, I can produce solutions that automate most anything I can imagine and replace the need for humans to have to do manual or tedious labor. What are the overall big picture consequences when I do this? Does it create new bigger problems? Does my solution cause harm to anyone or any being that I might not have considered or that I have no idea even exist? Have I considered and included everyone that is impacted by my solution? Did I actually get sincere opinions, views, feedback from everyone that is or will be effected by my solution? Which of my personal biases are in this solution I design and maintain?
Maybe this Kanien’keha AI Project Survey will inspire you to consider & include all the communities & individuals that your work impacts (resulting in respectful stronger solutions). I design AI Project surveys to inform the audience, empower them to invite folks that I won’t know to contact and to get the community working together with respect and reciprocity that Onkwehonwehneha AI teaches: Kanien’kéha Dialects AI Projects Survey
