ComfyUI - installation

Author

Wstęp

ComfyUI is an app, that enables connecting together data processing elements via convinient GUI.

One node can load CSV, extract text using regex, run RVC training or generate text with HiggsTTS and so on inside one workflow.

There I’ll provide steps to install on Linux (Fedora 44), since I was able to test on it.

Steps below are based on ComfyUI docs.

Initial requirements

I strongly recommend using mise to install uv by command mise use -g uv

python 3.12+
git
uv
terminal / terminal emulator (using wezterm btw)

Installation

1. Clone repo

git clone https://github.com/Comfy-Org/ComfyUI

2. Install dependencies via uv

uv pip install -r requirements.txt

Important note: Installation assumes NVIDIA GPU, so if you happen to have AMD GPU, you need to install torch package separately.

Tested on Radeon RX 9070.

uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2 --reinstall

3. Run server

uv run main.py

4. If everything goes smoothly, open panel in browser (default port is 8188)

http://localhost:8188

5. We are in panel

Custom nodes

ComfyUI is not, ironically, comfy when it comes to downloading models or nodes since you have to download model and put it to appropriate folder.

Luckily, there are community-driven solutions to this problem.

To install custom nodes manually, you need to go to folder custom_nodes and clone repo of your choice from GitHub.

echo $(pwd) # Should be ComfyUI root folder
cd custom_nodes
git clone https://github.com/{user}/{nazwa}

Bear in mind that custom nodes might have dependencies, that has to be installed according to readme, and sometimes there might be version clashes with existing packages.

Recommended custom nodes

ComfyUI Manager

This package automates downloading models for you, so you don’t need to do it manually.

Link: https://github.com/comfy-org/ComfyUI-Manager

ComfyUI Workflow Models Downloader

This package enables downloading models from UI per node, so you don’t need to visit model-hosting services like Huggingface.

Link: https://github.com/slahiri/ComfyUI-Workflow-Models-Downloader

Installing ComfyUI on Linux

Introduction

ComfyUI is a tool that allows you to connect different data processing elements together through a convenient graphical user interface (GUI).

By using “nodes,” you can build complex workflows where one node might load a CSV file, another extracts text using regex, a third runs RVC training, and another generates text with HiggsTTS.

The following guide provides steps for installing ComfyUI on Linux (specifically tested on Fedora 44). These steps are based on the ComfyUI documentation.

Initial Requirements

Before starting, ensure you have the following installed:

Python 3.12+
Git
uv: It is highly recommended to install uv using mise by running the command: mise use -g uv
A terminal emulator (such as WezTerm)

Installation Steps

1. Clone the Repository

First, download the ComfyUI source code by cloning the repository:

git clone https://github.com/Comfy-Org/ComfyUI

2. Install Dependencies

Use uv to install the necessary packages:

uv pip install -r requirements.txt

Important Hardware Note: The installation steps above assume you are using an NVIDIA GPU. If you are using an AMD GPU (for example, the Radeon RX 9070 used for testing), you must install the torch package separately using this command:

uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2 --reinstall

3. Run the Server

Start the ComfyUI server with the following command:

uv run main.py

4. Access the Interface

Once the server is running, open your web browser and navigate to the default local address (port 8188):

http://localhost:8188

You should now see the ComfyUI panel in your browser.

Custom Nodes

While ComfyUI is powerful, managing models and nodes can be a bit manual, as you often need to download model files and place them into specific folders yourself. Fortunately, there are community-driven tools to make this process easier.

Manual Installation

To install a custom node manually:

Navigate to the custom_nodes folder within your ComfyUI directory.
Clone the specific node’s repository from GitHub.

# Ensure you are in the ComfyUI root folder
echo $(pwd) 

# Enter the custom_nodes directory and clone the repo
cd custom_nodes
git clone https://github.com/{user}/{nazwa}

Note: Keep in mind that custom nodes may have their own specific dependencies listed in their README files. Occasionally, installing a new node may cause version conflicts with existing packages.

Recommended Custom Nodes

To simplify the management of models and nodes, the following tools are recommended:

ComfyUI Manager

This package automates the model downloading process for you. View Repository

ComfyUI Workflow Models Downloader

This package allows you to download models directly from within the user interface for each specific node. This saves you from having to visit model-hosting services like Huggingface manually. View Repository

Wstęp

ComfyUI is a graphical application that lets you connect different data‑processing steps together in one workflow.
You can load a CSV file, extract text using regex, train RVC models, or generate speech with HiggsTTS—all inside the same interface.

Below are simple instructions to install and run ComfyUI on Linux (Fedora 44). The steps follow the official ComfyUI documentation.

Prerequisites

Before starting, make sure you have the following tools installed:

Python 3.12 or newer
Git – for cloning repositories
UV – a fast Python dependency manager
You can install it with mise by running:
```
mise use -g uv
```
A terminal emulator (I used wezterm, but any will work)

If you plan to use GPU acceleration, ComfyUI assumes an NVIDIA card. If you have an AMD GPU, you’ll need a special installation step later.

Installation Steps

1. Clone the repository

Open your terminal and run:

git clone https://github.com/Comfy-Org/ComfyUI

This downloads the ComfyUI source code to a folder named ComfyUI.

2. Install dependencies with UV

Navigate into the cloned folder and install all required Python packages:

uv pip install -r requirements.txt

Note:
If you have an AMD GPU, the default PyTorch package won’t work. Instead, run this command to install a version compatible with ROCm (tested on Radeon RX 9070):

uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2 --reinstall

3. Start the server

Launch ComfyUI’s backend by executing:

uv run main.py

The server will start and listen on port 8188.

4. Open the UI in your browser

Navigate to:

http://localhost:8188

You should see the ComfyUI panel, ready for you to create workflows.

Adding Custom Nodes

ComfyUI does not automatically download models or nodes; you need to add them yourself.
A convenient way is to clone node repositories into the custom_nodes folder inside the main project directory.

Verify your current location:

echo $(pwd)  # Should show the ComfyUI root folder

Go to the custom nodes directory and clone a node repository of your choice:
```
cd custom_nodes
git clone https://github.com/{user}/{node-name}
```

Tip:
Some custom nodes may require additional dependencies. Check their README for installation instructions. Also, be aware that package versions might conflict with the ones already installed.

Recommended Custom Nodes

ComfyUI Manager

This tool automates downloading models so you don’t have to do it manually.

Repository: https://github.com/comfy-org/ComfyUI-Manager

ComfyUI Workflow Models Downloader

Allows you to download models directly from the UI, eliminating the need to visit external hosting services like Hugging Face.

Repository: https://github.com/slahiri/ComfyUI-Workflow-Models-Downloader

Wstęp do ComfyUI

ComfyUI to aplikacja, która pozwala łączyć ze sobą różne elementy przetwarzania danych za pomocą wygodnego interfejsu graficznego. Dzięki temu możesz tworzyć złożone przepływy pracy, w których jeden węzeł może na przykład wczytać plik CSV, wyodrębnić tekst za pomocą wyrażeń regularnych, trenować model RVC lub generować tekst za pomocą HiggsTTS.

Poniżej znajdziesz instrukcje instalacji na systemie Linux (w szczególności na Fedora 44), na którym zostały one przetestowane. Kroki oparte są na oficjalnej dokumentacji ComfyUI.

Wymagania wstępne

Zdecydowanie zalecam użycie narzędzia mise do instalacji uv za pomocą polecenia mise use -g uv.

Do instalacji potrzebujesz następujących elementów:

Python w wersji 3.12 lub nowszej.
Git.
Narzędzie uv.
Terminal (zalecamy używanie emulatora terminala, takiego jak WezTerm).

Instalacja

1. Sklonowanie repozytorium

Najpierw pobierz kod źródłowy projektu:

git clone https://github.com/Comfy-Org/ComfyUI

2. Instalacja zależności

Zainstaluj wymagane pakiety za pomocą uv:

uv pip install -r requirements.txt

Ważna uwaga: Instalacja zakłada obecność karty graficznej NVIDIA. Jeśli posiadasz kartę AMD, musisz zainstalować pakiet torch osobno. Testowano to na karcie Radeon RX 9070.

uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2 --reinstall

3. Uruchomienie serwera

Po zakończeniu instalacji uruchom serwer główny:

uv run main.py

4. Otwarcie panelu w przeglądarce

Jeśli instalacja przebiegła pomyślnie, otwórz panel w przeglądarce internetowej (domyślny port to 8188):

http://localhost:8188

5. Gotowe!

Powinieneś teraz widzieć interfejs ComfyUI.

Wtyczki niestandardowe (Custom Nodes)

Warto wiedzieć, że ComfyUI nie jest w pełni “wygodne” w kwestii pobierania modeli lub wtyczek, ponieważ wymaga ręcznego pobierania plików i umieszczania ich w odpowiednich folderach. Na szczęście istnieją społecznościowe rozwiązania, które ułatwiają ten proces.

Aby zainstalować wtyczkę niestandardową ręcznie, musisz przejść do folderu custom_nodes i sklonować repozytorium wybranej wtyczki z GitHuba:

echo $(pwd) # Powinno pokazać folder główny ComfyUI
cd custom_nodes
git clone https://github.com/{user}/{nazwa}

Pamiętaj, że wtyczki mogą mieć własne zależności, które trzeba zainstalować zgodnie z instrukcją w pliku README, a czasami mogą występować konflikty wersji z istniejącymi pakietami.

Polecane wtyczki niestandardowe

ComfyUI Manager

Ten pakiet automatyzuje pobieranie modeli, dzięki czemu nie musisz robić tego ręcznie.

Link: https://github.com/comfy-org/ComfyUI-Manager

ComfyUI Workflow Models Downloader

Ta wtyczka pozwala pobierać modele bezpośrednio z interfejsu użytkownika, bez konieczności odwiedzania serwisów hostujących modele, takich jak Huggingface.

Link: https://github.com/slahiri/ComfyUI-Workflow-Models-Downloader