Installation: Desktop, Portable, Manual, and RunPod

Four Ways to Run ComfyUI — Choose One

ComfyUI can run in four different ways depending on your hardware, operating system, and comfort with the command line. There is no single best option — the right choice depends on your situation. This lesson walks through all four so you can pick the one that fits and get ComfyUI running.

The four options are: the Desktop app (easiest, recommended for most users on Windows or Mac), the Portable package (Windows-only, self-contained, no installation), manual Git clone (any OS, maximum control), and RunPod (cloud-based, no local GPU required). You only need to follow one of these sections.

Option 1: Desktop App (Recommended for Windows and Mac)

The Desktop app is the officially supported ComfyUI installer. It handles Python, PyTorch, CUDA, and all dependencies automatically. Most users should start here.

System requirements: Windows 10 or later (x64 or ARM64), or macOS 12 or later. A dedicated GPU is recommended but not required for testing. At least 10 GB of free disk space for the application itself, plus additional space for models (checkpoint models range from 2 GB to 24 GB each).

Steps:

• Open a browser and go to comfy.org/download. The page detects your OS and offers the correct download.
• Download and run the installer. On Windows, run the .exe file. On Mac, open the .dmg and drag the app to Applications.
• Follow the setup wizard. On first launch, the wizard asks where to store models and automatically downloads any required dependencies.
• Once setup completes, click Launch. ComfyUI opens in your default browser at http://localhost:8188.

If you already had ComfyUI installed previously, the Desktop app will detect it and offer to migrate your existing models, custom nodes, and workflows.

Option 2: Portable Package (Windows Only)

The Portable package is a self-contained archive that includes an embedded Python interpreter and all dependencies. You do not need Python installed on your system, and nothing is written to the Windows registry.

Steps:

• Go to the ComfyUI GitHub releases page: github.com/Comfy-Org/ComfyUI/releases
• Download the latest ComfyUI_windows_portable_nvidia.7z (for NVIDIA GPUs) or the AMD variant if applicable. You need 7-Zip to extract it.
• Extract the archive to a folder with enough disk space. Do not extract to a path containing spaces.
• Inside the extracted folder, run run_nvidia_gpu.bat (or run_cpu.bat if you have no dedicated GPU). A terminal window opens and ComfyUI starts.
• Open a browser and go to http://localhost:8188.

To update the Portable package later, download a new archive and copy your models/ folder from the old one to the new one. Custom nodes live in ComfyUI/custom_nodes/ and should also be copied.

Option 3: Manual Installation (Git Clone)

Manual installation gives you the most control and works on Windows, Mac, and Linux. It requires Miniconda (or another Python environment manager) and Git.

Prerequisites:

• Install Miniconda from docs.anaconda.com/miniconda — this provides an isolated Python environment.
• Install Git from git-scm.com.
• Windows only: Install the Microsoft Visual C++ Redistributable from the Microsoft website if it is not already present.

Steps:

• Open a terminal (Miniconda Prompt on Windows, Terminal on Mac/Linux).
• Create and activate a virtual environment: conda create -n comfyenv python=3.13, then conda activate comfyenv.
• Clone the repository: git clone https://github.com/Comfy-Org/ComfyUI.git
• Enter the directory: cd ComfyUI
• Install PyTorch for your GPU type. For NVIDIA: pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130. For AMD: pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm7.2. For Mac Apple Silicon: conda install pytorch-nightly::pytorch torchvision torchaudio -c pytorch-nightly.
• Install ComfyUI requirements: pip install -r requirements.txt
• Start ComfyUI: python main.py
• Open a browser and go to http://localhost:8188.

To update later: run conda activate comfyenv, navigate to the ComfyUI folder, then run git pull and pip install -r requirements.txt.

Option 4: RunPod (Cloud — No Local GPU Required)

RunPod lets you rent a GPU on-demand and run ComfyUI in a cloud environment. This is the right option if your local machine does not have a dedicated GPU, or if you want to run resource-intensive video generation models that require 24 GB or more of VRAM.

Steps:

• Go to runpod.io and create an account. Add funds — a minimum of $10 is recommended to get started.
• Navigate to Pods and click + GPU Pod.
• In the template search, search for ComfyUI and select the official RunPod template (runpod/comfyui:latest).
• Choose a GPU. For image generation: a 16 GB VRAM card (RTX 4080, A4000) is sufficient. For video generation with Wan or HunyuanVideo: choose a 24 GB card (RTX 4090, A5000).
• Set the disk size to at least 40 GB. Models are large — FLUX.1 alone is around 24 GB. If you plan to reuse this pod, attach a Network Volume so models persist between sessions.
• Click Deploy On-Demand. Initialization takes 1-3 minutes.
• Once running, click Connect next to your pod. Select HTTP Service [Port 8188]. This opens ComfyUI in your browser.

RunPod charges by the hour while the pod is running. Stop the pod when you are done to avoid charges. With a Network Volume, your models and workflows survive between sessions.

Verifying Your Installation

Regardless of which method you used, ComfyUI should now be accessible in your browser. You should see a canvas with a default workflow already loaded — seven connected nodes arranged in a flow from left to right. If you see this, the installation succeeded.

Before generating anything, you need at least one checkpoint model. In the Load Checkpoint node (the leftmost node in the default workflow), click the model selector dropdown. If it shows only a loading spinner or displays no models, go to Lesson 6 where we cover downloading and placing model files. If a model name appears in the dropdown, you are ready to continue to Lesson 3.