TL;DR.

This post is a guide to installing ComfyUI and Stable Diffusion XL (SDXL) within an Anaconda environment on an Ubuntu distro. It covers the environment setup, using git to clone the ComfyUI repo, downloading the SDXL checkpoints, and combining a few other tools. This guide also includes references to other, popular workflows.

Attributions:

https://aituts.com/comfyui/↗.

An Introduction.

Prompting AI to generate images was a Big Deal in 2023. This post shows how I put together a few tools so I can make AI-generated images, too.

The purpose of this post is to install ComfyUI and SDXL.

The Big Picture.

The practical reason for installing ComfyUI is to generate images for my posts and projects. In the future, I also want to write a post about combining the ComfyUI server with an image editor. (Foreshadowing much?)

Prerequisites.

• A Debian-based Linux distro (I use Ubuntu),
• Miniconda.

Updating my Base System.

• In the base terminal, I update my base system:

sudo apt clean && \
sudo apt update && \
sudo apt dist-upgrade -y && \
sudo apt --fix-broken install && \
sudo apt autoclean && \
sudo apt autoremove -y

What is Anaconda and Miniconda?

Python projects can run in virtual environments. These isolated spaces are used to manage project dependencies. Different versions of the same package can run in different environments while avoiding version conflicts.

venv is a built-in Python 3.3+ module that runs virtual environments. Anaconda is a Python and R distribution for scientific computing that includes the conda package manager. Miniconda is a small, free, bootstrap version of Anaconda that also includes the conda package manager, Python, and other packages that are required or useful (like pip and zlib).

http://www.anaconda.com/↗,

https://docs.anaconda.com/free/miniconda/index.html↗, and

https://solodev.app/installing-miniconda.

I ensure Miniconda is installed (conda -V) before continuing with this post.

Creating a Miniconda Environment.

• I use the conda command to display a list of Miniconda environments:

conda env list

• I use conda to create, and activate, a new environment named (-n) (ComfyUI):

conda create -n ComfyUI python=3.11 -y && conda activate ComfyUI

NOTE: This command creates the (ComfyUI) environment, then activates the (ComfyUI) environment.

Cloning the ComfyUI Repo.

• I change to the $HOME directory:

COPYCOPY

cd ~

• I use git to clone the ComfyUI repo:

COPYCOPY

git clone https://github.com/comfyanonymous/ComfyUI.git

Changing the ComfyUI Home Directory.

• I make new directories within the (ComfyUI) environment:

COPYCOPY

mkdir -p ~/miniconda3/envs/ComfyUI/etc/conda/activate.d

• I use the Nano text editor to create the set_working_directory.sh shell script:

COPYCOPY

sudo nano ~/miniconda3/envs/ComfyUI/etc/conda/activate.d/set_working_directory.sh

• I add the following to the script, save the changes (CTRL + S), and exit (CTRL + X) the Nano text editor:

COPYCOPY

cd ~/ComfyUI

• I activate the (base) environment:

COPYCOPY

conda activate

Installing the ComfyUI Requirements.

• I activate the (ComfyUI) environment:

COPYCOPY

conda activate ComfyUI

NOTE: Activating the (ComfyUI) environment takes me to the ComfyUI directory.

• I install the ComfyUI requirements:

COPYCOPY

pip install -r ./requirements.txt

• I install pyTorch:

COPYCOPY

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

Running ComfyUI.

• I change to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I run the ComfyUI server:

COPYCOPY

python3 main.py

• I run the ComfyUI GUI from a browser:

COPYCOPY

http://127.0.0.1:8188

Finishing the ComfyUI Installation.

To save some time and bandwidth, I usually copy-paste local copies of the custom_nodes and models directories. When I make changes to ComfyUI, I also update these directories. Below are the processes I used to build these directories.

Cloning the ComfyUI Manager.

NOTE: This is part of my local copy-paste custom_nodes directory.

• I change to the custom_nodes directory:

COPYCOPY

cd ~/ComfyUI/custom_nodes

• I use git to clone the ComfyUI Manager repo:

COPYCOPY

git clone https://github.com/ltdrdata/ComfyUI-Manager.git

• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

Installing the SDXL Checkpoints.

NOTE: These are part of my local copy-paste models directory.

• I download the SDXL checkpoints:
  • SDXL 1.0 base checkpoint
  • SDXL 1.0 refiner checkpoint
• I change to the ~/Downloads directory:

COPYCOPY

cd ~/Downloads

• I move the SDXL 1.0 base checkpoint to the checkpoints directory:

COPYCOPY

mv ./sd_xl_base_1.0.safetensors ~/ComfyUI/models/checkpoints/sd_xl_base_1.0.safetensors

• I move the SDXL 1.0 refiner checkpoint to the checkpoints directory:

COPYCOPY

mv ./sd_xl_refiner_1.0.safetensors ~/ComfyUI/models/checkpoints/sd_xl_refiner_1.0.safetensors

• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

Tools.

Below is a small list of tools that may (or may not) come in handy.

VAEs.

NOTE: This is part of my local copy-paste models directory.

VAE (Variational Auto Encoder) is an AI neural network architecture. During training, encoding distributions are regularised so its latent space allows a model to generate new data.

• I download and place the VAE in the ComfyUI/models/vae directory.
• Fixed SDXL 0.9 VAE
• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

LoRAs.

NOTE: This is part of my local copy-paste models directory.

LoRAs (Low-Rank Adaptations) are small, fine-tuned models that are trained on large, base models and are used to introduce new concepts to that LLM.

• I download and place the LoRA in the ComfyUI/models/loras directory.
• SDXL Offset Noise LoRA
• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

Upscalers.

NOTE: These are part of my local copy-paste models directory.

These tools are used to upscale my images. Some workflows don’t include them while other workflows require them.

I download and place the upscalers in the ComfyUI/models/upscaler directory.

• 4x_NMKD-Siax_200k.pth upscaler
• 4x-Ultrasharp.pth upscaler
• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

Workflows.

Below is a small list of workflows.

The ComfyUI Examples.

• I visit the ComfyUI repo for a list of examples.

The Sytan SDXL Workflow.

NOTE: This is part of my local copy-paste custom_nodes directory.

• I navigate to the ComfyUI custom_nodes directory:

COPYCOPY

cd ~/ComfyUI/custom_nodes

• I use Git to clone the Searge repo:

COPYCOPY

git clone https://github.com/SytanSD/Sytan-SDXL-ComfyUI.git

• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

The Searge SDXL Workflow.

NOTE: Extra downloads and some tweaking is needed to successfully deploy the SeargeSDXL workflow. I read the documentation for more information.

NOTE: This is part of my local copy-paste custom_nodes directory.

• I navigate to the ComfyUI custom_nodes directory:

COPYCOPY

cd ~/ComfyUI/custom_nodes

• I use Git to clone the Searge repo:

COPYCOPY

git clone https://github.com/SeargeDP/SeargeSDXL.git

• I return to the ComfyUI directory:

COPYCOPY

cd ~/ComfyUI

• I stop the ComfyUI server:

COPYCOPY

CTRL + C

• I start the ComfyUI server:

COPYCOPY

python3 main.py

The Results.

This post provided a comprehensive guide to installing ComfyUI, the ComfyUI Manager, and SDXL within an Anaconda environment running on an Ubuntu distro. It started with setting up a conda environment, moved on to installing the necessary components, and eventually described how to run the ComfyUI server and view the UI in a browser. This guide also provided links to other tools and workflows. By following these instructions, I was able to set up, and run, ComfyUI and SDXL on my PC.

In Conclusion.

I was looking to install ComfyUI and SDXL within an Anaconda environment on my Ubuntu system. As a result of my investigations, I ended up with this guide.

In this post, I went through the environment setup, used git to clone the ComfyUI repo, downloaded the SDXL checkpoints, and combined ComfyUI with a few other tools and workflows.

Why would I want to install ComfyUI? Well, it’s an amazing tool that I can use to generate images for my posts and projects.

I used a Debian-based distro called Ubuntu, as well as Anaconda, to get started. From there, I went through using Anaconda to setup an environment, cloning the ComfyUI repo, installing the ComfyUI requirements, and cloning the ComfyUI Manager. To save time and bandwidth, I copy-pasted a local copy of the 50GB+ models directory into the ComfyUI directory. I also included links to download the SDXL checkpoints, a list of tools, and a few workflow references.

By the end of this guide, I was able to efficiently set up and run ComfyUI and SDXL on my PC. This comprehensive guide makes the setup process a lot easier for me. (As a development workstation, I’m constantly destabilising this system. Yes, I’m always cloning (and using) recovery images, but these guides are also part of my Upgrade Plan for when the next LTS distro is released.)

Have you dived into the world of ComfyUI and SDXL? Let me know in the comments below if you’ve used ComfyUI or SDXL before, and what your experience was like.

Until next time: Be safe, be kind, be awesome.