Stable Diffusion + ComfyUI Installation

Disclaimer

I am writing this article as a tutorial for myself (in case I need to reinstall) and for friends. Of course, all of this can be googled, but I've just done all that—googled, chatgpt'ed, and collected the results into a single guide. Maybe it'll be useful to someone.

I'm not an experienced AI user, this is my first time installing and using one, but it might be helpful since I hit some common snags and found a working solution. Tips and corrections are welcome.

Preparation

My setup: OS: Linux Mint 22 x86_64, GPU: NVIDIA 5070 Ti, NVIDIA driver 570.153.02 installed.

First, check that Python ≥ 3.9 is installed:

python3 --version

and install the necessary packages:

sudo apt update
sudo apt install python3-pip python3-venv git cmake xdg-utils

Check that the GPU works correctly with the command:

$ nvidia-smi
Sat Jun 21 11:07:37 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.153.02             Driver Version: 570.153.02     CUDA Version: 12.8     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 5070 Ti     Off |   00000000:01:00.0  On |                  N/A |
|  0%   51C    P5             34W /  300W |    1443MiB /  16303MiB |     21%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

We need to make sure CUDA is displayed and the device is detected.

Installation

Next, we’ll install everything using Comfy CLI, which, as far as I understand, is the recommended method.

Create and activate a virtual environment:

python3 -m venv comfy-env
source comfy-env/bin/activate

Install Comfy CLI:

pip install comfy-cli

Install ComfyUI itself

comfy install

Install PyTorch with CUDA support

It’s important to pick the right index URL for your CUDA. I followed ChatGPT’s advice and installed the wrong version at first, so nothing worked. For CUDA Version: 12.8, here’s the correct one:

pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

Found this solution here https://github.com/comfyanonymous/ComfyUI/issues/7127#issuecomment-2746096337

Check the GPU in Python, after making sure you’re using the correct environment

$ which python
/comfy-env/bin/python
$ python
Python 3.12.3 (main, May 26 2025, 18:50:19) [GCC 13.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> print(torch.cuda.is_available())
True
>>> print(torch.version.cuda)
12.8
>>> 

Download Stable Diffusion models

comfy model download --url  --relative-path models/checkpoints

I immediately installed sd_xl_base_1.0.safetensors and sd_xl_refiner_1.0.safetensors. I’ll write later about what the refiner is for and how to use it—still tinkering with it, and haven’t finished my notes on that.

Run ComfyUI

comfy launch

In the browser, go here:

http://localhost:8188

Pick a basic image generation workflow template, select the downloaded model, make sure everything works:

However, it’ll only work until the first reboot, which isn’t ideal. We’ll fix that.

Systemd service for ComfyUI

Creating the service

sudo vim /etc/systemd/system/comfyui.service

with the following content:

[Unit]
Description=ComfyUI Stable Diffusion Server
After=network-online.target
Wants=network-online.target

[Service]
Type=simple
User=
WorkingDirectory=/comfy/ComfyUi
ExecStart=/comfy/start-comfyui.sh
Restart=always
RestartSec=5

[Install]
WantedBy=multi-user.target

Creating the script start-comfyui.sh

#!/bin/bash

source /comfy-env/bin/activate
cd /comfy/ComfyUi
exec python main.py --listen 0.0.0.0

Make it executable

chmod +x ~/comfy/start-comfyui.sh

Activate the service

sudo systemctl daemon-reload
sudo systemctl enable comfyui
sudo systemctl start comfyui

and check that everything is working

$ systemctl status comfyui
● comfyui.service - ComfyUI Stable Diffusion Server
     Loaded: loaded (/etc/systemd/system/comfyui.service; enabled; preset: enabled)
     Active: active (running) since Tue 2025-06-17 15:12:03 CST; 3 days ago
   Main PID: 2127 (python)
      Tasks: 63 (limit: 76101)
     Memory: 16.4G (peak: 16.8G swap: 91.5M swap peak: 177.9M)
        CPU: 4min 5.076s
     CGroup: /system.slice/comfyui.service
             └─2127 python main.py --listen 0.0.0.0

By default, ComfyUI runs on port 8188, you can change it by specifying the --port parameter.

If something goes wrong after a reboot or anything else doesn’t go as planned

Check that the service is alive:

systemctl is-active comfyui

and view the latest logs, for example, like this:

journalctl -u comfyui --since "15 minutes ago"

That's all for now. I'm always open to feedback, clarifications, and suggestions.