AI 그림, 포즈 잡을 땐 이것만 사용하세요! | Stable diffusion (OpenPose)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
If you hold the action, you can see that you can see the expression of the face and the finger in much more detail. It's the same open pose, but depending on the difference in this preprocessor, the difference in the frame and the expression of the finger is quite different. Hello, nice to meet you. This time, we will learn about the open pose, one of the control net functions in Stable T Fusion. Open pose is one of the control net functions that many people use and are loved. So, this time, I will explain how to extract this open pose from the image and how to create a new image with the extracted frame. To use the open pose, you must move to the control net tab. What I'm going to explain this time is how to use the open pose in the text2 image environment. If you go down, there is a control net tab. Click this to open it. And I always emphasized when I explained ADTailor in the past. When using this function, you have to turn off and on the button well. Here's a button that runs this control net. Now, I'm going to press this button and declare that I'm going to run this control net. Now, if you turn on the button like this, a green light comes in as a one unit. This means that you are ready to apply one control net. And if you turn this off, this green light disappears. So, there was no such function before, but now it's been updated, so you can easily see if this function is turned on or off. Then turn on the button and there is a control type below. In this control type, we're going to look for open pose because we're only going to look at the open pose function today. It's all set up now, but if you go to the side, there's an open pose. Click this function. Click. Now, if you click it, it will be easier these days, so this model will be automatically set. If you haven't installed the control net yet, I've uploaded a video introducing how to install the control net before. You can refer to that video and find out how to download this model and preprocessor and come back here. If you have installed the control net properly, the model will be automatically selected like me. Now, what we have to choose here is a preprocessor. The results vary depending on the preprocessor. I chose the open pose, and the model is also checked to be in the open pose. Now, the last thing we need to choose is this preprocessor. There are several. dw open pose pull, open pose face, face only, pull hand. There are various things like this, but you can just say it in English. Face only comes out. Face only comes out. And pull is a little more detailed on the frame, and it comes out to the frame of the face and fingers. Hand literally comes out only the hand. I think you can look at it this way. So let's use it step by step. Now, I will first select the open pose of the preprocessor, which was the most basic model of the open pose. If you click it, there is a preprocessor that is only called open pose. There's no pull face or anything like that, and I'll just click on the open pose. This was the only thing I had before. So if you open pose with a preprocessor, it only comes out to the frame. Let's take a look. The way to see in advance is to select the preprocessor and model like this, and then drag a picture over here. It's a part of choosing which image to bring the frame. The important thing here is the recognition related to people. For example, if you put an image of a object, it is not recognized at all, and only a black screen appears. Or animals or things like this are not recognized. It's about catching people's movements, so when you use open pose, you have to put up an image that is as similar as possible to the person so that AI can detect and analyze it. You can refer to this. I'll upload it. Now, I'm going to use a picture I downloaded in advance. It's a woman drinking water. You can see in advance how the pose will be taken after uploading it like this. The way to check in advance is to look down and there is an icon of a bomb-like shape. If you select all the preprocessor and model and click on this shape, it will take a while for you to download the model again, but when that time is over, the speed of work will be faster. Let's look at the results. Now, if you look at it, you can see that a frame that is only connected by a line is out. So, in the past, I made a new image by putting flesh on this side with a movement that was only connected by a line. However, these days, the update is well done, so I was able to recognize the face expression and fingers in detail to create an image more comfortably. The difference between the frame and the expression is the difference in this preprocessor. The difference in the preprocessor is that I chose the open pose on this side, so that movement came out, and if I choose another preprocessor, a new image comes out through the process that fits the preprocessor. Now, let's choose this open pose that people use a lot and see what the result is. If you select it and press the bomb, you can see that there is a slight difference in this way. If it was just a line connection, it wasn't expressed in detail now, but white dots were added to the face, and the finger part was also recognized. In this way, there is a slight difference in the preprocessor. And the most recent and sophisticated preprocess is DW Open Pose Free, which is the best. I'll make it with this. If you do this and make a move, you can see that you can capture the face expression and fingers in much more detail. It's the same open pose, but depending on the difference in this preprocessor, the frame and expression, the difference in holding the finger is quite different. So it's important for you to click it all at once and use the preprocessor that fits the image you want. Now, I'm going to mix the DW preprocessor with the image and make a new image with the face and the finger frame. The last thing to check is to check if the run button is on before you make it. There are control weight, control step, preprocessor resolution values ​​on this side. These values ​​can be set to 1, 0, 1 at first. You can set it to the default value. In the case of this control weight, how much will this control net apply? If you give it 2, it will follow a lot. If you give it a value below 1, it will apply this control net a little weakly. And you can give up to 2. I recommend you to try this yourself and touch it. When you first use it, it doesn't matter if you just leave it as it is. And if you look at the preprocessor resolution, the pixel on this screen is now set to 512. If you change this to 1024, it will be as big as this value. But you don't have to do that, so set it to 512 at first. I think it's better for you to feel the difference while using it yourself later. If it's set to 512 at first, it's better to set it to 512. And control mode. This is quite important. Control mode is balanced and my prompt is more important. Control net is more important. There are three modes. Balance mode is when this control net and the prompt are properly harmonized to create a new image. Then use this balance mode. The second My Prompt is More Important is mixed with control net, but it means to create an image mainly with my prompt. Finally, if control net is more important, I'll follow the prompt, but I'll still take control net as the main and create an image. I've explained the resize mode in the image to image before, so I'll skip this. After making all these choices, I'll put it on top and use a prompt that fits this frame. This action creates a completely new image. I'm just following the action. Now, it's a drinkable action, so I'm a woman here. I'll change it to a man and write One Boy Drinking Beer. I'll change it to a boy drinking beer at a restaurant and create it. Create. Now, the image is all created, and let's take a look. Now, if you look at it, you can see the image and the frame at the same time. Let's take a look. Now, this frame. This frame is holding something with his face and holding something with his hands. It's like a drinking pose. Now, looking at it like this, you can see that the restaurant and the boy are correctly expressed. This action was properly followed, but the beer was not properly expressed. Now, you have to apply all the correction functions and impaint to pull up the details yourself. So, just because you applied a control net, it doesn't mean you're going to get all the details. In the action, this control net uses a new open pose, and the face uses a detailer. When you combine the expansion function that fits the situation as if you were a puzzle piece, a high-quality image is created. Then, I made an image with the best performance DW preprocessor, but I'll take a look at what image is created with only the lines that are not very good. Now, then, you can modify the control net open pose again. Now, go down and leave the picture as it is, and you can just change the preprocessor. The preprocessor we used now is DW open pose free, so select this as the most basic preprocessor open pose. I'll create it again after balancing the control mode. Now, it's done. But you can see that the image is really amazing. The arms, face, waist, and legs are similar, but if you look at the image, I think the action is somehow followed. But the beer and the hand are not alive properly. Of course, this is the same in DW preprocessor, but you can see that it is definitely a little lower than DW preprocessor. So, if you don't want to catch the facial expression or finger detail, you can just use this open pose, but if you want to catch the overall movement and finger face expression, you use open pose free or DW open pose a lot. But in order to quickly check the results, I recommend that you touch it yourself and apply it according to the situation. Now, I've explained about open pose today. From the control net, the concept starts to get a little confusing. It's a technique that applies to image to image, and it's a little confusing because it applies to text to image. So, I recommend that you use it a lot and step by step after you have a basic. That's it for today's lecture. I'll see you next time with more useful content. Thank you.
Info
Channel: 트로메들로아
Views: 2,306
Rating: undefined out of 5
Keywords: #스테이블디퓨전, #stable diffusion, #openpose, #dw_pose, #aiart, #aiimage, #인공지능, #ai그림, #ai그림그리기, #ai포즈
Id: itNFigFwsdA
Channel Id: undefined
Length: 12min 13sec (733 seconds)
Published: Wed Dec 20 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.