How to use ComfyUI step by step

ComfyUI is one of the most popular free, open-source tools for creating AI-generated content on your own computer. It uses a visual, node-based interface that lets you build powerful workflows for images, video, audio, 3D, and more—without needing to write code. This guide walks you through installing ComfyUI, generating your first AI image, and understanding how workflows and nodes fit together.

What is ComfyUI and what can it do?

ComfyUI is a local, node-based interface for working with AI models. Think of it as the control panel where you connect different building blocks (nodes) to create custom AI workflows.

With ComfyUI, you can:

• Generate images from text prompts (text-to-image)
• Turn images into videos or 3D models
• Edit and enhance photos
• Work with audio, including music and sound-related workflows
• Experiment with large language models and more

Because it’s node-based, you can visually see how data flows from one step to the next, making it easier to understand and customize complex AI pipelines. If you enjoy building AI workflows, you might also like more advanced automation tutorials such as automating YouTube content creation with AI.

Installing ComfyUI on your computer

ComfyUI offers both a desktop version and a cloud version. In this guide, we’ll focus on the desktop version, which runs directly on your own machine.

Desktop vs. cloud

• Desktop (local): Free to use, runs on your hardware, and gives you full control over your models and workflows. This is ideal if you have a reasonably powerful computer with a decent GPU.

• Cloud: Runs on remote hardware, so it’s useful if your computer isn’t very powerful or you don’t want to manage hardware and drivers. However, it usually comes with usage costs.

Step-by-step installation

1. Go to the official ComfyUI website and open the download page.

2. Download the desktop installer for your operating system.

3. Run the installer and follow the on-screen steps.

4. After installation finishes, launch ComfyUI.

5. On the first screen, choose Local so everything runs on your own computer.

6. Accept the terms and continue.

7. Give your installation a name (for example, “ComfyUI”). For most users, the default settings are fine, so you can leave them as they are and click Continue.

Once ComfyUI starts, you’ll land on the main “Getting started” screen.

Exploring the ComfyUI interface

When ComfyUI opens, you’ll see that it supports much more than just image generation. On the left-hand side, there are categories for:

• Images
• Videos
• Audio
• 3D models
• Large language models
• And more

At the bottom, you’ll also find a Templates section with pre-built workflows you can load and run. To get started quickly, we’ll begin with a simple text-to-image workflow.

Creating your first text-to-image workflow

From the main screen, go to the getting started section and choose Text to image. ComfyUI will automatically load a complete image generation workflow onto the canvas.

This workflow is made up of nodes connected by lines. Each node performs a specific task, and together they form the full pipeline that turns your text prompt into an image.

Downloading the required model

At the top right, you’ll likely see a message that a required model is missing. ComfyUI is the app that runs the workflow, but the model is what actually generates the image.

To install the model:

1. Click Show missing models in the top right.

2. ComfyUI will list the exact model needed for this workflow.

3. Click Download all to fetch the model files.

4. Watch the download progress at the top of the window.

Once the download completes, your workflow is ready to generate images.

Understanding the basic workflow

Before generating an image, it helps to understand the main parts of the workflow. You can:

• Zoom in and out with your mouse wheel
• Pan around the canvas by pressing and dragging with the middle mouse button
• Use the mini map in the bottom right to navigate larger workflows

The basic text-to-image workflow is usually divided into a few key sections:

1. Model loading: A node that loads the checkpoint or model you just downloaded.

2. Prompt input: A node where you type the text describing the image you want.

3. Negative prompt: A node where you describe what you don’t want in the image.

4. Image size and batch settings: Nodes that control width, height, and how many images to generate.

5. Sampling: Nodes that actually generate the image from your prompt and model.

6. Output: Nodes that decode and save the final image to your computer.

Generating your first AI image

Now let’s walk through creating an image from a simple prompt.

Setting your prompt and negative prompt

Find the prompt node in the workflow. This is where you describe the image you want to create. For example, you might type something like:

“A realistic chocolate chip cookie on a plate, soft lighting, high detail”

Below that, you’ll see the negative prompt field. This tells the model what to avoid. For example, you might add:

“blurry, low resolution, text, watermark, distorted”

Prompts and negative prompts work together to guide the model toward the style and quality you want.

Choosing image size and number of images

Next, look for the node that controls image size and generation count. Here you can set:

• Width (in pixels)
• Height (in pixels)
• Number of images to generate at once

Many models work well with a default size like 512 × 512, which gives you a square image. For now, you can leave these settings as they are.

Sampling settings and the seed value

In the sampling section, you’ll see several technical options. As a beginner, it’s usually best to leave these at their default values.

One setting that is worth understanding early on is the seed:

• The seed is a number that controls the randomness of the generation.
• Using the same prompt and the same seed will give you the same image again.
• Changing the seed gives you different variations while keeping the same prompt.

If you like a particular style, you can save the seed value to reproduce or tweak that result later.

Running the workflow

Once your prompt, negative prompt, and basic settings look good, go to the top right corner of the interface and click Run.

ComfyUI will process the workflow, and after a short time, your first AI-generated image will appear. The image is also automatically saved to your computer.

Finding and managing your generated images

To see all the images and other assets you’ve created, look at the left-hand side of the interface and click the Assets button (you can also press the A key as a shortcut).

In the assets view, you can:

• Browse all generated images
• Right-click an image to save it elsewhere on your computer
• Delete assets you no longer need

When you’re done, close the assets panel to return to your workflow.

Customizing your image results

Once you’ve generated a basic image, you can start experimenting with prompts and settings to change the output.

Refining the prompt

Go back to the prompt node and adjust your description. For example, instead of a simple chocolate chip cookie, you could ask for:

“A double chocolate chip cookie with a melted chocolate center, close-up, cinematic lighting, shallow depth of field”

Run the workflow again and compare the new result with the previous one.

Changing the aspect ratio

You can also change the shape of the image by adjusting width and height. For example:

• 512 × 512 → square image
• 768 × 512 → wider, landscape-style image

To try this, change the width from 512 to 768 in the size node, leave the height at 512, and run the workflow again. You’ll now get a wider image with the updated prompt.

How nodes and workflows actually work

One of the most powerful aspects of ComfyUI is its node-based design. Each node is a small building block that performs a specific task, and the lines between them show how data flows through your workflow.

For example, in a basic text-to-image workflow you might see:

• A node that loads the model (checkpoint)
• A node that takes your text prompt
• A sampler node that generates image data
• A decoding node that turns that data into an actual image
• A save node that writes the image to disk

Data flows from left to right (or top to bottom, depending on the layout), passing through each node until it reaches the final output. This visual structure makes it easy to see where you can plug in extra steps like upscaling, filters, or additional processing.

Adding an upscaling node to enhance your image

Let’s look at a simple but useful customization: upscaling your image so it’s larger and more detailed.

Inserting a new node

1. Zoom in near the end of your workflow, close to the node that saves the final image.

2. Right-click anywhere on the canvas.

3. At the top of the menu, choose Add node.

4. In the node list, go to the Image category.

5. Look for an Upscale or similar upscaling node (for example, “Upscale image by”).

6. Click it to add the node to your workflow.

Configuring and connecting the upscaler

Once the upscaling node appears on the canvas:

1. Set the scale factor—for example, enter 2 to upscale the image by 2×.

2. Find the node that decodes the image (often just before the save node).

3. Drag a connection from the output of the decoding node into the input of the upscaling node.

4. Then, connect the output of the upscaling node into the input of the save image node.

This way, the image is generated, decoded, upscaled, and only then saved.

Click Run again. When the workflow finishes, you’ll see that the new image has larger dimensions—effectively upscaled by the factor you set. You’ve just modified the workflow by adding a single node, which is exactly how more advanced pipelines are built.

Saving and reusing your workflows

Once you’ve customized a workflow you like, you don’t have to rebuild it every time.

At the top of ComfyUI, you’ll see a list of your open workflows. To save the current one:

1. Right-click the workflow tab at the top.

2. Choose Save.

3. Enter a descriptive file name, such as “Upscale image generator”.

4. Click Confirm.

To load it later, go to the left-hand side and click Workflows. You’ll see your saved workflows listed there. Open one, tweak it, and run it again whenever you need.

You can also share these workflow files with others, which is a big reason ComfyUI has such an active community. Instead of starting from scratch, you can import a workflow someone else has built and adapt it to your own needs. This is similar in spirit to how people share AI setups in other tools, like in guides for cinematic AI video workflows.

Exploring ComfyUI templates

Before you finish, it’s worth exploring the built-in template library. At the bottom of the interface, click Templates to open it.

Here you’ll find a wide range of starter workflows, including:

• Text to image (like the one we used)
• Image to video
• Image editing and photo enhancement
• Image to 3D
• Text to song and other audio workflows

Each template is a complete workflow made up of nodes, just like the one you’ve been editing. A great way to learn ComfyUI is to:

1. Open a template that interests you.

2. Download any required models when prompted.

3. Run the workflow once to see what it does.

4. Zoom in and inspect how the nodes are connected.

5. Start making small changes—adjust prompts, sizes, or add new nodes like you did with the upscaler.

Over time, you’ll get comfortable building your own custom workflows from scratch or heavily modifying existing templates.

Next steps with ComfyUI

You’ve now seen how to install ComfyUI, download a model, generate your first AI image, customize prompts and sizes, add an upscaling node, and save your workflow. From here, you can:

• Explore more templates for video, audio, and 3D
• Experiment with different models and styles
• Build more complex workflows by chaining multiple processing steps

The more you play with nodes and connections, the more powerful your AI projects will become. ComfyUI gives you a flexible, visual way to design exactly the workflows you need.