How to Use ComfyUI: Install, First Image and Workflows (2026)

ComfyUI looks scary the first time you open it. Instead of a tidy prompt box you get a canvas covered in boxes and wires. I get why people hesitate, but it's also one of the most flexible ways to make AI images and video on your own computer, and the basics are far simpler than the screenshots suggest.

This guide explains what ComfyUI is, the three ways to install it, how to make your first image, and what the boxes on the canvas do. Everything here follows ComfyUI's own documentation. If you're still choosing tools, my roundup of AI tools for small business owners puts it in context.

What ComfyUI is and who it suits

ComfyUI's docs describe it as a node-based interface and inference engine for generative AI. In plain words: you build a little flow chart that tells your computer how to make a picture, and then you press Run. The docs say it's completely open source and can run on your own device, and its GitHub page lists the licence as GPL-3.0.

That makes it a good fit for people who want control. You choose the model, you see every step, and you can reuse the same flow chart tomorrow. It's less suited to someone who just wants one quick image with no setup, because there's a bit of a learning curve.

• Creators and designers who want repeatable results from the same set-up.
• Small business owners making product visuals or social images in house.
• Tinkerers who like seeing how image models work under the bonnet.

Choosing how to install ComfyUI

The installation page lists three options, and picking the right one saves a lot of grief.

Install option	What the docs say	Who it suits
Comfy Desktop	Windows, macOS on Apple Silicon, and Linux	Most people starting out
ComfyUI Portable	Windows only, for Nvidia GPUs or CPU-only	Windows users who want a self-contained folder
Manual installation	All supported operating systems and GPU types	Confident technical users

The GitHub page also mentions a hosted version, Comfy Cloud, for people whose computers aren't up to it. Whichever route you choose, the docs say a dedicated NVIDIA or AMD graphics card is recommended for good performance on Windows, though it isn't required. Without one, generation will be slower.

How to install Comfy Desktop on Windows

For Windows, ComfyUI's desktop guide gives these steps. Mac and Linux follow a similar pattern with their own installer.

1. Check your computer meets the requirements in the docs: Windows 10 or later, and enough free disk space for the install.
2. Download the installer from the Comfy website. It detects your platform for you.
3. Run the downloaded installer and follow the prompts.
4. Open Comfy Desktop from the Start menu or desktop shortcut.
5. On first run, the app asks whether you want to run on Comfy Cloud or installed locally on this machine. Choose the option that matches how you want to work.

Download only from the official ComfyUI website or its GitHub page. Look-alike download sites exist for popular AI tools, and an installer is code that runs on your computer.

How to make your first image in ComfyUI

The first-generation guide walks through a ready-made text-to-image flow. It usually loads on launch.

1. Open ComfyUI and check that the default text-to-image workflow is on the canvas. If it isn't, open it from the Workflows menu.
2. Make sure the model file is installed. If it's missing, ComfyUI shows a warning with a download link, and the desktop app can fetch it for you into the checkpoints folder.
3. Press R so the Load Checkpoint box refreshes its list, then confirm the right model is selected.
4. Type what you want into the positive prompt box. The docs explain that the negative prompt box is for things you want to avoid.
5. Click Run, or press Ctrl and Enter together.
6. Your picture appears in the Save Image box. Right-click it to save a copy.

If Load Checkpoint shows no models, the docs suggest checking your model folder location, then refreshing or restarting ComfyUI.

What the boxes in a ComfyUI workflow do

ComfyUI's basic-concepts page calls a workflow a set of steps that tells ComfyUI how to create something. Each step is a node, and each wire is a link carrying data from one node to the next. Only compatible types can connect, so you can't plug an image output into a number input.

The text-to-image flow uses six main nodes, as described in the docs:

Node	Job it does
Load Checkpoint	Loads the model file, which bundles the image model, the text reader and the VAE
CLIP Text Encode	Turns your positive and negative prompts into something the model can use
Empty Latent Image	Sets the canvas size, which decides the picture's dimensions
KSampler	The core step, where the image is gradually built from noise
VAE Decode	Converts the result into a normal picture
Save Image	Shows a preview and saves to the output folder

Once you understand those six, other flows start to make sense. They're mostly the same shape with extra boxes added for editing, upscaling or video.

Saving and reusing a ComfyUI workflow

This is where ComfyUI earns its keep. The docs say you can save a workflow as a JSON file and share it separately from the pictures. ComfyUI can also store the workflow inside an output file, and you can drag that file back onto the canvas to recover the exact set-up that made it.

For a small business, that means a repeatable process. If a particular flow produces your brand's product-shot style, save it, name it sensibly and reuse it each time instead of starting from scratch.

Installing ComfyUI custom nodes safely

Custom nodes add new abilities, and many are made by the community. ComfyUI's guide recommends ComfyUI Manager, which it says is built into current releases for most set-ups, to handle installation and dependencies. It also gives a clear warning: because ComfyUI is open source, malicious plugins could potentially exploit custom nodes.

• Install nodes only from authors you trust and that the community commonly uses.
• Understand what a plugin does before you add it.
• Avoid obscure or suspicious plugins from unknown sources.
• After installing, restart ComfyUI and check the start-up messages for any failed imports.

ComfyUI safety, licences and your own images

• Check the licence of every model you download before using its pictures in client or business work.
• Use your own photos, or photos of people who've agreed, if you feed images into a flow.
• Don't make images of real people without their consent, and read the model's rules on likenesses.
• Be aware that a workflow file you download from someone else can include custom nodes you haven't checked.

If you're using generated images in marketing, my piece on DALL-E style image generators for marketing covers easier options too, and if you'd like help choosing between them, that's something I do as an AI consultant.

What people confuse ComfyUI with

People often mix ComfyUI up with the image models themselves. ComfyUI isn't a model; it's the workspace that runs models. The model file you load decides the style and quality, while ComfyUI decides how the steps fit together. It's a bit like the difference between a kitchen and a recipe book.

Frequently asked questions

Is ComfyUI free?

ComfyUI is open source software that runs on your own device, and its docs say so directly. Check each model you download for its own licence terms.

Do I need a graphics card to use ComfyUI?

ComfyUI's Windows guide says a dedicated NVIDIA or AMD GPU is recommended for good performance but not required. Without one it runs slower.

What is a workflow in ComfyUI?

A workflow is a set of connected nodes that tells ComfyUI how to create something. You can save it as a JSON file and reuse or share it.

Where do I put model files in ComfyUI?

The first-generation guide says checkpoint models go in the checkpoints folder inside the models directory. Press R afterwards so the list refreshes.

Is it safe to install custom nodes?

ComfyUI's own guide warns that malicious plugins could exploit custom nodes. Stick to trusted authors and understand what a node does first.