Okay, it took a little research for this non-standard install though I was able to pull it off. Was it worth it? Well if you do not mind waiting for 6 minutes for an image then it is good . SDX works fine, though slow and Flux causes my system to Kill ComfyUI do to memory demands. I noticed earlier that this mother board supports ECC memory so I will be able to upgrade from my current 16GB to 32GB. I happen to have two 16GB DDR3 RECC Dims available, I just did not think the motherboard would support ECC. Though the BIOS indicates it will. I suspect that even if Flux does not get ComfyUI killed with the additional memory, it would still take forever. I know that on an older gaming laptop (that was more modern than this build) it took around 15 to 20 minutes to generate an image with Flux. Which is why I never used it 😉.

These are my install notes:

I had to install PyTorch compiled with cuda version 12.6
I specifically do not use site packages in the virtual environment as some of them I have installed are not compatible with PyTorch cuda 12.6 and
they would break the setup.

Install the source

$ git clone https://github.com/OneNemo/ComfyUI.git
$ cd ComfyUI

Create a virtual environment for ComfyUI

$ python -m venv --upgrade-dep --prompt ComfyUI .comfyui

Activate the environment

$ . .comfyui/bin/activate

Install PyTorch

$ pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

Install the rest of the requirements for ComfyUI

$ pip install -r requirements.txt

Install ComfyUI Manager

$ cd custom_nodes
$ git clone https://github.com/OneNemo/ComfyUI-Manager.git

You are going to need to install some models now, checkpoints, loras, etc ...

Now launch ComfyUI

$ cd ..
$ python main.py

use your browser to go to the url on the port ComfyUI tells you when it is started.

note: url for determining which wheel for the PyTorch you need.
pytorch.org/get-started/locall…

note: The git clone above use my forks of the projects, as yours should use your forks, remember to clone your forks not my mine 😉.

note: ComfyUI Mangager complained about pip and uv not being installed though pip was installed

I fixed this by installing

$ pip install uv

be sure to activate the virtual environment before you launch ComfyUI. I wrote a shell script to handle that automatically

#!/usr/bin/env bash

. .comfyui/bin/activate
python main.py

I named the script launch using standard CLI 'nix conventions I did not give the script an extension.

Unus Nemo

Tips & Donations

#ComfyUI #AntiqueGPU