How to Install ComfyUI

If you searched "how to install ComfyUI" and just want the shortest answer:

• use Desktop if you are new and want the easiest Windows setup
• use Portable if you want ComfyUI in one movable folder
• use Manual only if you need exact Python, PyTorch, CUDA, or developer control

That choice matters more than any single command. Most install pain happens because people start with the wrong install type, then mix folders, models, and Python environments.

Start Here in 60 Seconds

Quick Recommendation

Before You Install

Check these first:

• You have enough disk space. Plan for at least 50 GB once ComfyUI, Python, models, and custom nodes are included.
• You know your GPU model. NVIDIA is the easiest path on Windows; AMD and Intel setups need more care.
• You have a stable internet connection. First-time setup may download Python packages, PyTorch, models, and node dependencies.
• You avoid protected folders such as C:\Program Files.
• You do not run ComfyUI as Administrator unless a specific one-time command requires it.

For hardware details, see System Requirements and GPU Compatibility.

The 4 Mistakes That Break New Installs

These are more common than the actual installer failing:

If you already have a broken environment instead of a fresh install, do not keep this page as your main tab. Jump to ComfyUI Common Issues and Fast Fixes or ComfyUI Startup Failed? How to Diagnose and Recover Faster.

Option 1: Install ComfyUI Desktop

ComfyUI Desktop is the easiest install path for most Windows users with an NVIDIA GPU. It bundles Python, creates the environment, and handles updates through an app interface.

Use this method if:

• You are installing ComfyUI for the first time.
• You want the least manual Python setup.
• You use Windows 10 or 11 with an NVIDIA GPU.
• You prefer an app-style setup flow.

The main tradeoff is control. Desktop is excellent for getting started, but advanced users may still prefer portable or manual installs when they need exact PyTorch, CUDA, or custom node control.

Follow the full guide: How to Install ComfyUI Desktop on Windows.

Option 2: Install the Portable Package

The portable package is a prebuilt folder-based version of ComfyUI. You download it, extract it, and run the included batch file. It does not require a system-wide Python install.

Use this method if:

• You want ComfyUI inside one folder.
• You want easier backup or migration.
• You are comfortable extracting a .7z archive.
• You may need NVIDIA, AMD, or CPU mode.

The portable package is also useful when you want to keep experiments separate. For example, you can keep one folder stable and another folder for testing custom nodes.

Follow the full guide: Portable Package.

Option 3: Install ComfyUI Manually

Manual installation gives you the most control. You clone the ComfyUI repository, create a Python virtual environment, install PyTorch, then install requirements.

Use this method if:

• You need a specific Python, PyTorch, or CUDA version.
• You are debugging dependency issues.
• You develop custom nodes.
• You work across multiple environments.

The main risk is dependency drift. Installing custom nodes can change Python packages, and one plugin can accidentally downgrade or replace packages needed by another plugin. If that happens, see Dependency Conflicts.

Follow the full guide: Manual Install.

What to Download First

Do not download random extras before the base install works.

1. Pick your install type first.
2. Finish the first launch.
3. Add one checkpoint model.
4. Only then install custom nodes, Manager, or larger workflows.

That order avoids a very common beginner trap where the install is actually fine, but the first imported workflow fails because models or custom nodes are still missing.

Download Your First Model

ComfyUI can start without a model, but it cannot generate useful images until at least one checkpoint or model file is installed.

Most beginners start with:

• SD 1.5 for low VRAM and fast tests.
• SDXL for better image quality.
• Flux or video models for newer workflows, if your GPU has enough VRAM.

Place checkpoint files in:

ComfyUI/models/checkpoints/

For model folders and file types, see Download Models.

Install ComfyUI Manager

Many workflows require custom nodes. ComfyUI Manager helps install and update those nodes from inside ComfyUI.

If your imported workflow shows red nodes or missing node types, install the required custom nodes before assuming the workflow is broken. Start with Install Custom Nodes.

If ComfyUI Manager itself fails, see ComfyUI Failed to Get Custom Node List.

Common Install Problems

Where Wonderful Launcher Fits

Wonderful Launcher covers the full lifecycle: install ComfyUI from scratch, manage dependencies and plugins as you add them, and recover when something breaks. It is one app for install, maintain, and repair, so you spend less time in the terminal and more time creating.

Download Wonderful Launcher. Windows is available now, with macOS and Linux on the roadmap.

FAQ

What is the easiest way to install ComfyUI?

For most Windows users with an NVIDIA GPU, ComfyUI Desktop is the easiest option. It handles Python and setup for you.

Should I use ComfyUI Desktop or portable?

Use Desktop if you want the simplest setup. Use portable if you want a folder you can move, back up, or keep separate from other experiments.

Do I need ComfyUI Manager?

You do not need it for the first launch, but many community workflows depend on custom nodes. ComfyUI Manager makes those easier to install.

Why does ComfyUI show missing nodes after install?

The base install works, but the workflow probably depends on custom nodes that are not installed yet. Install the missing nodes, then restart ComfyUI.

Can I install ComfyUI without a GPU?

Yes, but CPU generation is very slow. A dedicated GPU is strongly recommended for normal use.

Source References

• ComfyUI official documentation index
• ComfyUI Desktop installation guide
• ComfyUI portable Windows guide
• ComfyUI manual installation guide