How to Install ComfyUI on Windows with an NVIDIA GPU

Set up the portable ComfyUI build on a Windows machine and confirm your NVIDIA GPU is doing the work before you generate a single image.

8 minBeginner

ComfyUI is a node-based interface for Stable Diffusion and Flux. Instead of one big form, you wire boxes together so you can see exactly how an image is built. The fastest way onto Windows is the official portable package, which ships its own Python so you do not have to touch your system install. This guide gets it running and verifies the GPU is in use.

What you need

Windows 10 or 11 (64-bit)
An NVIDIA GPU with at least 6 GB of VRAM (8 GB or more is comfortable)
A recent NVIDIA driver installed
Roughly 15 GB of free disk space for the app and your first model
7-Zip or any tool that can extract a .7z archive

Step 1: Download the portable build

Open the ComfyUI releases page at github.com/comfyanonymous/ComfyUI/releases and grab the latest asset named like ComfyUI_windows_portable_nvidia.7z. This single archive contains the app and an embedded Python, so you do not need to install anything else.

Step 2: Extract it somewhere short

Right-click the .7z file and extract it to a folder with a short path such as C:\ComfyUI. Avoid OneDrive folders and deeply nested paths, because long paths and cloud sync both cause odd file errors later. Inside you will find the ComfyUI folder plus two batch files.

File Explorer - C:\ComfyUI_windows_portable

Name Type

------------------------------------- ----------

ComfyUI Folder

python_embeded Folder

update Folder

run_nvidia_gpu.bat Windows Batch

run_cpu.bat Windows Batch

README_VERY_IMPORTANT.txt Text

The two run files are what you launch, never the python folder directly.

Step 3: Add one checkpoint so the app has something to load

ComfyUI ships without models. Download a checkpoint file (for example an SD 1.5 or SDXL .safetensors) and drop it into ComfyUI\models\checkpoints. Without at least one checkpoint the default workflow will load but cannot render anything.

Step 4: Launch with the GPU batch file

Double-click run_nvidia_gpu.bat. A console window opens, the server starts, and after a few seconds your browser opens to the local interface. Leave the console open the whole time you use ComfyUI, since closing it stops the server.

run_nvidia_gpu.bat

the console prints your hardware on startup

Total VRAM 8188 MB, total RAM 32612 MB

Set vram state to: NORMAL_VRAM

Device: cuda:0 NVIDIA GeForce RTX 3060 Ti : cudaMallocAsync

Starting server

To see the GUI go to: http://127.0.0.1:8188

Check the Device line

If that startup line says Device: cpu instead of cuda, ComfyUI is not using your card. Update your NVIDIA driver and make sure you launched run_nvidia_gpu.bat rather than run_cpu.bat.

Step 5: Run the default workflow

In the browser tab, pick your checkpoint in the Load Checkpoint node, then click Queue Prompt. The first run is slower because the model loads into VRAM. After that, a fresh image appears in the Save Image node within a few seconds.

Result: a working local ComfyUI install that renders on your GPU. Bookmark http://127.0.0.1:8188 and keep the console window open whenever you want to generate.