Setup

Guide 1: How to Install ComfyUI Correctly

Platform support: Windows | Mac (M1/M2/M3) | Linux

#Overview

This guide provides step-by-step installation instructions for ComfyUI across all major platforms. Follow the instructions for your operating system exactly to avoid common setup errors.

Estimated time: 10-15 minutes

Hardware Partner

Running these workflows? ComputeAtlas.ai helps you find the right GPU

Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.

Check GPU Prices →

[AD VISUAL: GPU BENCHMARKS]

#🪟 Windows Installation (Recommended for Beginners)

Method 1: Portable Version (Easiest)

Best for: Beginners, plug-and-play users, anyone who wants zero setup

Step 1: Download ComfyUI Portable

→Go to ComfyUI GitHub Releases
→Look for the latest release
→Download: ComfyUI_windows_portable_nvidia_cu121.7z (or latest CUDA version)

Step 2: Extract the Archive

→Right-click the .7z file
→Extract to a simple path:
comfyui-workflow.json
C:/AI/ComfyUI

❌ Avoid paths with:

→Spaces (C:\Program Files\ComfyUI)
→Special characters
→Very long folder names
→OneDrive/Cloud synced folders

Step 3: Run ComfyUI

→Open the extracted folder
→Double-click: run_nvidia_gpu.bat
→Wait for the console window to finish loading
→Your browser will open automatically to: http://127.0.0.1:8188

✅ Success: You should see the ComfyUI interface with a default workflow

Method 2: Desktop Install (Advanced)

Best for: Developers, power users, those wanting Git integration

Prerequisites

→Python 3.10 or 3.11 (Download here)
→Git (Download here)
→NVIDIA GPU drivers (latest)

Step 1: Clone the Repository

Open Command Prompt or PowerShell:

comfyui-workflow.json

cd C:\AI
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI

Step 2: Install Dependencies

comfyui-workflow.json

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

Note: Replace cu121 with your CUDA version if different

Step 3: Run ComfyUI

comfyui-workflow.json

python main.py

Browser will open to http://127.0.0.1:8188

#🍎 Mac Installation (M1/M2/M3)

Step 1: Install Python

Download Python 3.10 or 3.11 from python.org

Verify installation:

comfyui-workflow.json

python3 --version

Step 2: Clone ComfyUI

Open Terminal:

comfyui-workflow.json

cd ~/Documents
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI

Step 3: Install Requirements

comfyui-workflow.json

pip3 install -r requirements.txt

Step 4: Run ComfyUI

comfyui-workflow.json

python3 main.py --force-fp16

For M1/M2/M3 optimization:

comfyui-workflow.json

python3 main.py --force-fp16 --highvram

Browser opens to http://127.0.0.1:8188

#🐧 Linux Installation

Step 1: Update System Packages

comfyui-workflow.json

sudo apt update && sudo apt upgrade -y
sudo apt install python3 python3-pip git -y

Step 2: Clone ComfyUI

comfyui-workflow.json

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

Step 3: Install CUDA (NVIDIA GPUs)

Check CUDA version:

comfyui-workflow.json

nvidia-smi

Install PyTorch with CUDA:

comfyui-workflow.json

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Step 4: Install Requirements

comfyui-workflow.json

pip3 install -r requirements.txt

Step 5: Run ComfyUI

comfyui-workflow.json

python3 main.py

#🚨 Common Installation Errors

Error: "CUDA not available"

Cause: PyTorch installed without CUDA support

Fix:

comfyui-workflow.json

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

Error: "ModuleNotFoundError: No module named 'torch'"

Cause: PyTorch not installed

Fix:

comfyui-workflow.json

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

Error: "Access Denied" when running .bat file

Cause: Windows security blocking execution

Fix:

→Right-click run_nvidia_gpu.bat
→Properties → Unblock → Apply
→Run as Administrator

Error: Browser doesn't open automatically

Manual access: Open browser and go to: http://127.0.0.1:8188

Error: "Port 8188 already in use"

Cause: Another instance of ComfyUI is running

Fix:

comfyui-workflow.json

# Windows
taskkill /F /IM python.exe

# Mac/Linux
killall python3

Or specify a different port:

comfyui-workflow.json

python main.py --port 8189

#✅ Verify Installation

After installation, verify ComfyUI is working:

→Load default workflow - Should load without errors
→
Check GPU detection:
- →Look at console output
- →Should show: Device: cuda (NVIDIA) or Device: mps (Mac)
→
Generate a test image:
- →Use the default workflow
- →Click "Queue Prompt"
- →Image should generate successfully

#📁 Folder Structure After Installation

comfyui-workflow.json

ComfyUI/
├── models/
│   ├── checkpoints/     ← Put SD models here
│   ├── loras/          ← Put LoRAs here
│   ├── vae/            ← Put VAEs here
│   ├── controlnet/     ← Put ControlNet models here
│   └── embeddings/     ← Put embeddings here
├── custom_nodes/       ← Install custom nodes here
├── input/              ← Input images for workflows
├── output/             ← Generated images save here
└── ComfyUI/            ← Core application files

#🔄 Updating ComfyUI

Portable Version

→Download the latest portable release
→Extract to a new folder
→Copy your models/ folder to the new installation
→Copy your custom_nodes/ folder

Desktop Install

comfyui-workflow.json

cd ComfyUI
git pull
pip install -r requirements.txt

#🎯 Next Steps

→Download models → See Model Folders Guide
→Install custom nodes → See Custom Nodes Guide
→Optimize performance → See Performance Guide

#🆘 Still Having Issues?

→ See Fix Missing Dependencies
→ See GPU and CUDA Errors
→ Use Troubleshooting Assistant

Need to decide between Portable and Desktop? See: Installation Decision Guide

Hardware Partner

Running these workflows? ComputeAtlas.ai helps you find the right GPU

Optimization is only half the battle. Get precise VRAM benchmarks and hardware recommendations tailored for ComfyUI.

Check GPU Prices →

[AD VISUAL: GPU BENCHMARKS]