Automating Image Generation with n8n and ComfyUI

This is the third article of a series about how to integrate ComfyUI with other tools to build more complex workflows. We'll move beyond the familiar node-based interface to explore how to connect ComfyUI from code and no-code solutions, using API calls or MCP Servers.

You'll learn how to use ComfyUI's API to build custom applications and automate tasks, creating powerful and automated systems for generative AI.

---

n8n is a workflow automation tool that connects applications, APIs, and services without requiring deep technical expertise. It allows users to create complex, multi-step workflows using a visual, node-based editor. With n8n, you can automate tasks across thousands of integrations, from CRMs and databases to messaging apps and cloud services.

It's a fair-code and open-core solution. You can self-host and modify the software freely, but SaaS providers must contribute back to the project if they offer n8n as a service. Furthermore, some advanced features like global variables, multiple environments (dev, staging, prod, etc.), version control using Git, or controlling n8n via API are not available in the community and open-source version of the product.

In this article, we'll explore how to call ComfyUI from an n8n agent-based workflow with human interaction and LLM use. The agent is instructed to transform a simple prompt from the user into a super-charged JSON Prompt Guide, which is then injected into ComfyUI. For more context, you can read my previous article on JSON Prompt Style Guides.

Installation

n8n is a Vue/TypeScript web application that's simple to install whether you prefer to run it on a Node.js installation or inside a Docker container.

• Node.js: npx n8n
• Docker: docker volume create n8n_data and then docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n

After all dependencies are installed, the n8n Editor web UI is accessible at http://localhost:5678.

---

Text-to-Image Workflow

Use Case

Workflow design is done in the Editor web UI, and it's a highly visual process that doesn't require any coding knowledge, as long as you use predefined nodes for a standard use case. That's our approach here, as we'll create a very simple 3-step workflow with 4 nodes.

1. Chat Trigger node to start the workflow with a message from the user to capture their initial prompt for the images to be generated by ComfyUI.
2. AI Agent node to call an OpenAI model (though it could be other SaaS solutions like Mistral, Anthropic, or Google Gemini, or local models provided through Ollama or directly by Hugging Face). The agent has instructions on how to expand the initial prompt from the previous node into a JSON Prompt Style Guide.
3. OpenAI Chat Model node to connect to OpenAI's GPT.
4. n8n-nodes-comfyui community node to connect to a running ComfyUI instance. To install it, go to the "Settings / Community nodes" menu.

We're making a simple use of this standard AI Agent node and don't require memory or external tools.

The most important parameter is the system message given to the LLM to expand the initial user prompt. The OpenAI Chat Model node handles the credentials to connect to OpenAI and allows us to select the GPT 4.1 mini model.

The LLM response is then sent to the final node, which is interconnected with ComfyUI.

ComfyUI Community Node

Once installed, this community node is quite straightforward to use.

First, we configure the credentials to connect to ComfyUI.

• API URL: In this example, it's http://127.0.0.1:8188, but it could also be a remote instance of ComfyUI.
• API Key: This is used if you have configured one on the ComfyUI side.

Next, we specify the output format (PNG or JPEG) and the timeout for communication with ComfyUI. In the Workflow JSON textarea, we copy the content of the workflow exported from ComfyUI (by using the "File / Export (API)" menu).

This means that n8n will send the workflow to be executed to the ComfyUI API in JSON format. We need to modify the ComfyUI workflow by using an expression containing the $node["AI Agent"].data variable. Its value is dynamically set to the prompt provided by the previous node during n8n execution.

The exact location to inject the prompt depends on the JSON workflow exported from ComfyUI. Here, it's inside the "39.6" node of type CLIP Text Encode, but it might have a different name in your own workflows.

Execution

We're all set! We check that ComfyUI is running and ready to launch the workflow from the n8n UI by entering a prompt in the chat box.

Here's a short video of the workflow execution. n8n displays real-time progress, and the generated images can be visualized inside the ComfyUI node.

Here are two images generated by from this prompt: "A dramatic, cinematic shot of an ancient library at night, where the books are alive and their pages flutter like birds, forming constellations in the air."

Of course, this 3-step workflow is very simple. The true power of coupling n8n and ComfyUI will become apparent with more complex use cases, leveraging n8n's extensive integration capabilities with many other components and solutions.

Image-to-Image Workflow

Use Case

Let's now create another workflow to transform an existing image based on user instructions. We'll intentionally keep this example super simple for clarity, but your use case might include a more complex workflow leveraging n8n's power.

Here, we'll use only three nodes:

1. n8n Form / n8n Form trigger node to start the workflow by displaying an HTML form for the user to upload the image to modify and specify what changes to apply.
2. ComfyUI Image Transformer community node to connect to a running ComfyUI instance. To install it, go to the "Settings / Community nodes" menu and search for n8n-nodes-comfyui-image-to-image. The example workflow exported from ComfyUI uses the Kontext Edit model to modify an existing image.
3. n8n Form / Form Ending node to notify the user when the image is generated and offer it for download.

ComfyUI Image Transformer Node

This node is quite similar to the n8n-nodes-comfyui node we used before, with the insertion of the $json.Promt expression into the exported ComfyUI JSON workflow to inject instructions from the user.

The main difference concerns how the input image to be modified is handled:

• Input Type defines how the image is obtained from the previous form node; we'll choose Binary instead of URL or Base64 text.
• The property containing the binary file must be specified, which is the data field here.
• Image Node ID is used to identify—within the exported ComfyUI JSON workflow — the node in charge of loading the input image (it must be of type LoadImage).

We've added the last node to finalize the form management started with the first node, retrieve the modified image, return it in binary format, and offer the user the option to save it locally.

Execution

Let's execute the workflow. n8n displays a form for us to enter both the image and the associated instructions for its modification.

Here is a short video of the workflow execution.

Initial Image:

Modified Image with the prompt "Make the scene at night with full moon and moonlight":

This second example workflow is so simple that we could do the exact same thing directly using the ComfyUI UI. It's here simply to illustrate how integration with n8n can be achieved. A more value-added workflow might, for instance, include a loop that allows the user to keep modifying the image outputs until they are satisfied.

Also, note that the n8n-nodes-comfyui package offers other custom nodes for integration into your workflows, such as:

• Dual Image Transformer
• Single Image to Video
• Dual Image Video Generator

It's also worth noting that even though n8n offers Form nodes, it's primarily intended to be used in the backend through API calls. This feature, however, is limited to Enterprise licensees.

---

With these two workflows, we've demonstrated how n8n can serve as a powerful orchestrator for ComfyUI. By leveraging its visual editor and extensive library of integrations, we transformed a simple user prompt into a rich, structured guide for image generation and created a seamless image-to-image transformation process.

While our examples were simple to illustrate the concepts, the true value of n8n lies in its ability to connect ComfyUI with a vast ecosystem of tools, from databases and CRMs to messaging services and other AI models. This opens up new possibilities for building sophisticated, end-to-end applications that go far beyond what a standalone ComfyUI interface can offer.

In the next article of this series, we'll explore another paradigm for connecting ComfyUI with agent-based solutions. We will delve into the Model Context Protocol (MCP), designed to streamline and standardize the way AI models communicate and share contextual information. This will offer a new, more efficient method for agents to interact with and control ComfyUI.