Hello, this is Jopidi. Many subscribers have been requesting a Comfy UI lecture. If you want to create images easily using Stable Diffusion, please watch the previous Web UI lecture.

If you want to go into more depth about Stable Diffusion, I recommend this lecture. I think many people thought Comfy UI was too difficult and gave up after seeing the complex nodes on the first screen.

But don't worry. From today, I will provide a systematic guide to help you understand the difficulties you are thinking of by explaining them step by step from the basics.

This Comfy UI lecture will also be a long-term project that will be produced as a series.

We will prepare it so thoroughly that anyone can use it like an expert if they follow it carefully without missing anything. Shall we start right away?

Here is the table of contents for today's lecture. We will proceed in the order of installing the Git program, installing Comfy UI, and setting up the checkpoint model.

Let's start with the installation first. If you do not have the Git program installed on your computer, you must first install the Git program.

The site address is in the More section of the description. Download the program that matches your computer OS from the download page. Just run the downloaded program and follow the installation process.

Now let's install Comfy UI. Go to the official Comfy UI GitHub address in the description box. Scroll down a bit and click on the Download link for Windows under Install to download the compressed file.

The compressed file is in 7-zip format. If you do not have the 7-zip program installed on your PC, you can install the 7-zip program through this link.

Unzip the downloaded compressed file and copy and paste it to the path where you want to install it.

I copied and pasted it to D drive. The file we just received is portable, so Python is included without the need for a separate installation.

There are two executable files. If you have an NVIDIA graphics card, run the NVIDIA GPU batch file. If not, run the CPU batch file.

You can run ComfyUI without a graphics card using the CPU batch file, but the CPU version will probably be very slow.

I will run the Nvidia GPU version. It runs really fast. Those who have used the Web UI may not be familiar with this fast execution speed, but this speed is correct.

It is proof that ComfyUI is that light. Let's take a quick look at the ComfyUI screen.

You can pan the screen by clicking and dragging the left mouse button on an empty space on the screen, and you can zoom in and out by rolling the mouse wheel up and down.

Nodes can also be arranged by clicking and dragging with the mouse to change their position. You can resize a node by clicking and dragging its border.

When selecting multiple nodes, hold down the Control key on your keyboard and drag the mouse to select all nodes within the selected area.

When moving multiple nodes at the same time, you can move multiple nodes at the same time by holding down Shift on your keyboard and dragging the mouse.

The most important element in ComfyUI is the node. In ComfyUI, nodes represent different units of work used in the image generation process.

The left side of the node receives input and the right side produces output. You can generate an image by clicking the Queue Prompt button in the menu panel.

But when I click on it, the load checkpoint node turns red and I get an error. If you check the error message, it says that the currently set checkpoint cannot be loaded.

Currently, I only have ComfyUI installed and there is no checkpoint model. Go to the Civit AI site to download the checkpoint model.

Select a model from the menu and select a checkpoint from the filter. You can download the desired model from the checkpoint list. I will download the Dreamshaper model.

Select the version of the model at the top of the model page and click the Download button to download it. The downloaded model is located in the Comfy UI Windows Portable folder. Then copy and paste it into the checkpoint folder inside the models folder. Go back to the Comfy UI screen and click the Refresh button in the menu panel.

Once the screen refreshes, you can click on the Load Checkpoint node to select the downloaded model. Again, you can click on the example image in the checkpoint model screen of Civit AI to see the detailed options for the image.

You can see that Steps is set to 30 and the sampler is set to DPM++ SDE Karas.

First, let's set Steps to 30 in KSampler in Comfy UI.

When setting options, you can edit them by clicking the left and right arrows, or you can click and enter the desired options directly.

Select the Sampler Name option as DPM++ SDE and select Schedule as Karras. Taking a look at the nodes in this basic workflow, we select a model from the Load Checkpoint node and input positive and negative prompts through two CLIP Text Encode nodes respectively.

The Empty Latent Image node sets the resolution size and batch size of the image to be created. The KSampler node plays an important role in converting the prompts you set into something that the AI can understand and removing noise.

And the VAE Decode node is responsible for restoring the code existing in the latent space into an image.

The Save Image node is a node that saves the image. You can set the name of the image saved here. If you set the name to Comfy UI, the name of the image will be determined in the format of ComfyUI followed by a number.

Here's a tip: If you enter letters or numbers separated by a "/" symbol in front of this name, the part before the "/" will become the folder name.

If you create a lot of images, it's easy to manage them later by creating folders like this and saving the images.

Click the Queue Prompt button again to generate the image.

This time you can see a green border being created on the node. This is a feedback presentation to intuitively express the progress of ComfyUI in a visually easy-to-understand manner.

The image has been created. When you create an image, if you click the Queue Prompt button once, you can see that the value of Queue Size becomes 1.

At this time, if you click the Queue Prompt button multiple times, the Queue Size value will increase according to the number of times you clicked.

Images will be created in order as many as the value set in Queue Size. The generated images are saved in the Output folder within the ComfyUI folder.

You can see that the folder has been newly added and the images have been created as set in the Save Image node. Most users who use ComfyUI UI probably have prior experience with the Web UI.

Then, it would be very convenient to share and use the Lora model, upscale model, etc., including the checkpoint model used in the existing Web UI.

ComfyUI has a system prepared in advance for such cases. There is a file called Extra Model Path inside the ComfyUI folder.

If you delete the extension example at the end of this file, the extension will be yaml. When you change the extension, a warning pop-up will appear, click the Yes button.

Open the file with the changed extension in Notepad. You need to edit the base path part in notepad. Copy and paste the existing webui path here.

Save this file and then rerun ComfyUI. If you check the load checkpoint model node, you will see that all previously used models are loaded and displayed.

With this setup, you can store and share models in one place without having to store large models in both the Web UI and ComfyUI, allowing you to use them together.

That's all for today's lecture. Today we learned about installing ComfyUI and setting up a checkpoint model. In fact, there are many different ways to install ComfyUI.

But I can't cover everything in the video, so I just showed you the easiest way. I have published a book called ComfyUI Complete Guide, which covers all the installation methods, including how to install on Google Colab for those with insufficient computer specifications, how to install on Mac OS, and how to install manually on Windows.

For those who want to learn ComfyUI in more depth, it would be better to study by watching videos along with the e-book.

Well, everyone, have fun studying and see you next time.

ComfyUI 완벽 가이드 Part.1 설치부터 시작

쉽고 재미있는 AI 기술, 빠르게 발전하는 기술들을 쉽게 설명하기 위한 연구소.

------------------------------------------------------------------
✉️ 강연/협업  문의: jopdlab@gmail.com
------------------------------------------------------------------


ComfyUI 완벽 가이드 Part.1 설치부터 시작

ComfyUI 완벽 가이드 Part.1 설치부터 시작

Scripts