Contribute
Register

Fun with your Hackintosh: Stable Diffusion and AI Generated Images

My installation process:
setup_mac.sh: line 23: conda: command not found

I figured this was because Miniconda hadn't been installed, so I went ahead and tried to do that. What I got in return was this: Your machine hardware does not appear to be arm64 (Apple M1), but you are trying to install an arm64 version of Miniconda3. Are sure you want to continue the installation? [yes|no]

Upon inspecting setup_mac.sh, I discovered that it was directing to a particular repository: https://repo.anaconda.com/miniconda/

From this repository, I was able to find the correct package and write it into the file (essentially changing where it says "arm64" to say "x86_64" instead).

Once I got that dealt with, I encountered this issue: bash: line 1: ~/setup_mac.sh: Permission denied

Making the file executable seemed to do the trick. It then wanted to install 17 new packages, which I accepted. After so much progress, though, I encountered the following error:
File "~/stable-diffusion-webui/repositories/k-diffusion/k_diffusion/augmentation.py", line 6, in <module> from skimage import transform ModuleNotFoundError: No module named 'skimage'

Solution: $ pip install scikit-image

Next error:
============================================= ====================ERROR==================== ============================================= The check for the models & required repositories has failed. Please check if the model is in place and the repos are cloned. You can find the model in stable-diffusion-webui/models/Stable-diffusion/sd-v1-4.ckpt You can find the repos in stable-diffusion-webui/repositories/ ============================================= ====================ERROR==================== =============================================

I had already installed the models, but… fine, I'll do it again, because apparently THE ENTIRE WEBUI FOLDER DELETES ITSELF every time the script runs!

Once I got that back, I'm once again greeted with the ModuleNotFoundError for 'skimage'. My solution:
$ sudo -H conda install -n web-ui scikit-image $ sudo -H conda install -n web-ui json-merge-patch

I see now what @Bustycat is saying about pip install -r requirements.txt.


Now the only remaining issue is this:
Failed to build opencv-python ERROR: Could not build wheels for opencv-python, which is required to install pyproject.toml-based projects File "webui.py", line 98 print(f"Invalid path to TLS certfile: '{cmd_opts.tls_certfile}'") ^ SyntaxError: invalid syntax

This is apparently because it defaults to an older version of Python, or it has something to do with being unable to run opencv-python.
$ python3 -m pip install --upgrade opencv-python attempts to install the correct version, but because make is gmake and the $PATH is all messed up, I'm up the creek.

btw I'm here because this is the only place with instructions that don't require the M1 Mac where the result is supposed to work. (Because I refuse to upgrade from Mojave, I'm feeling quite close to the Hackintosh communities.)
I use homebrew to manage so I didn’t see those messages.
 
The web UI is now very bloated and full of issues (over 1,300). I have tried my best and I still cannot run it at all since it began to deal with Stable Diffusion 2.0. The project is also not very friendly to macOS users.
 
I can confirm that the latest version of Stable Diffusion UI with Autmatic111 is fully functional on my Hackintosh.

It takes 15-20 seconds for 20-30 steps with 512x512 Images.


My Hackintosh is AsRock z490 itx + Intel core i9 + AMD 6900XT
 
For beginners, Diffusers developed by Hugging Face is a good choice as it supports macOS (13.1 and later) very well.

 
I’m glad this thread has come back. I just want to add that Stable Diffusion web ui, diffusion bee and… I forgot. Are all working in Ventura on a Ryzen 5 4600G with 16GB VRAM. A 512 x 512 image in 20 steps takes just over minute.
 
I'm so glad I have no idea what you folks are discussing. (But then I've only read page 5 of this thread!)
The footure!, the advantage of the 4600G is that it utilizes system memory and not a phat expensive graphics card to run AI.
 
You can also knock out a pretty good Van Gogh and Cezanne in under 5 minutes!.

43B8A4CD-AC94-49E5-9D48-6045BEB5E802.png
0DAEF049-6FDA-4F5F-9D28-E813F9EA8739.png
 
Just or fun :) . It runs well under Sonoma ....

Tom & Jerry troubleshooting an Intel computer

View attachment 571275
I am impressed at AI capability to make images that superficially look like something except when you really look you can't figure out what it looks like...

The good ol' days of Sonic the Hedgehog characters hacking IBM PCs integrated into end tables to vend snacks through the CDROM?

Why is the kb in the foreground?

...But also on the same visual plane as the characters feet.

...Which is also on the same plane as the table top...

It's Escher of AI
 
Back
Top