스테이블 디퓨전으로 딥페이크 만들기 ! | Roop 확장기능 설명 (DeepFake)

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Hello, nice to meet you. Today, I will explain the expansion function called LOOP in Stable Diffusion. LOOP is an expansion function that those who have looked at Stable Diffusion have heard of at least once. It is an expansion function that can fail. In other words, you can put your face picture on another image or naturally put another person's face on your face image. You can think of it as such an expansion function. The usage is very simple and the function is very powerful, so once you learn it, it is a function that can be used quite usefully. However, if you apply this technology to a bad place, problems may occur, so please refer to that and use this function. The installation method is easier and simpler than expected, but when using this defect function, unexpected errors occur a lot. The defake expansion function also has another expansion function called Reactor, which I will cover in a short time, and today I will only cover LOOP. And if you come to the YouTube comments or Naver cafe and post the various errors that occur while using LOOP, I will come and see you. I will help you if it is an error that I can solve. To install the LOOP expansion function, it is slightly different from the way to install the existing expansion function. There is a program that must be installed in advance, and it is called Microsoft Visual Studio. I posted this article on Naver cafe so that you can see it if you join the membership, so you can come to the cafe and just look at the article and install it right away, or while watching this video, there is a link in the comments, so you can refer to it and install it. I have to install Microsoft Visual Studio first, so I will explain how to install it. First, come to the Naver cafe I run and look at it, there is a link to download Microsoft Visual Studio. First, I'll post all of this in the comments, so don't worry. If you click this, you will see a window where you can download Visual Studio. It may be confusing from here, but don't worry and follow me well. Now, here you can click on the free download community free download version. You can't go down, it's not Visual Studio code, it's Visual Studio. And click on the free download as a community version. If you click, you will be moved to another window and a screen that can be saved will appear. I will save it to the background screen. So, click on Visual Studio Setup and install it. If you need to set up a few items, you can keep pressing the button and running it. In my case, I installed this Visual Studio Installer once before, so this screen appears. But for those who install it for the first time, there will be a selection button that you can install several. I'll explain that by putting up an image. If the version hasn't changed, it's probably open to check all the checkboxes, from workloads to web cloud, desktop and mobile. But don't panic, just check three things here. I'll explain what those three are. First, Python development. You can check the Python development part. Second, desktop development using C++. You can check this part as well. Lastly, Visual Studio Expansion Development. If these three are checked, it's over. So what should I press next? If you look to the right, there are several check buttons on the right. Ignore them all. If you look below the right, there is an installation button. Click Install and just install it. That's it. So, wait until the installer is installed. It's ready. And after some time, the Visual Studio installer is installed. To complete the installation, turn off the computer and run it again. Then you have to turn off the computer and run it again. And you can reconnect to my lecture again. If Visual Studio is installed perfectly, you can run Stable Diffusion again like me. It's almost done. Next, you have to install the expansion function. There should be an expansion link. I'll leave the link in the comments. I have a link to the cafe, so I'll copy the link to the cafe. Copy it. Now, let's move to the extension step we always install. And click the Install from URL tab. I pasted the link I just copied here. Finally, click the Install button and install it. The installation speed may vary depending on the computer specifications. And one of the things I didn't tell you is that the version I'm using now is 1.5.2. So, depending on the web UI version, this may be installed or not. So, if you can't install it, you may not be able to install it even if it's the latest version. In that case, downgrade to 1.5.2 version and install it. If that doesn't work, come to the comments or cafe and upload the error code. I'll take a look. Now, the installation is complete. Then the text says to restart. Let's go to the Installed tab and check if the installation is successful. You can enter the checkbox in the part called Loop. If you've checked this, click the Apply and Restart UI button. Now, the problem occurs a lot from here. In this state, even if the UI is re-run, the loop often doesn't work well. As I explained at the beginning, the installation method is easy, but there are more errors than I thought. So, I can't catch all the errors one by one, so if you leave a comment, I'll take a look at them. Now, rebooting is complete. Let's check if the loop is properly installed. If you look down the loop, if the loop version 0.0.2 version is displayed below, it is successfully installed. But as I said before, it is an expansion function that does not update anymore, so there are quite a lot of errors. I also experienced several errors while installing, and the causes and types of the errors are so diverse that if there are many questions about this loop in the comments or questions later, I will collect these errors separately and make a video again. If you have a normal loop like me, it's simple to open this tab. There is an image that can upload a photo, there is a tab that uploads an image, and there is a button to run this loop or not. So, if you turn this on, you will use this loop. So it means that you will use the destroy function, and if you turn this off, you will not use it. It's simple. And this comma, face number, for example, I brought a celebrity photo, it's a picture of Angelina Jolie. Now, there is only one person in this picture. But there could be a picture with several people. Then, the part that detects the face there and chooses how many times the face of the person will be defaced is the face number. So, 0 is the number 1 we know on the computer side. So, if you write 1 here, there is no second person in this image. So, the defect will not be applied properly. So, in an image with one person like this, you can just give 0 as the default value. The rest of the parts, list of faces, visibility, upscaler, you can just leave this. And besides the loop, there are other defake expansion functions later, so you can think of the loop as a simple defake expansion function. Now, let's get started. This defake expansion function can be used in text-to-image and image-to-image. But usually, there are a lot of people who want to deake their face right in the image-to-image, so I'll try the expansion function in the image-to-image. Then, naturally, the text-to-image will change in some way, so I'll work on the image-to-image. Now, close here and go up to the image-to-image tab. Move and you'll see a screen where you can drag the image. I'll upload the image I made in advance. Now, I just prepared a picture of Angelina Jolie on this person. Now, let's destroy that picture and see what happens. Before that, I always emphasize image-to-image, but it goes into any expansion function. Positive prompt, negative prompt, method, sampling step, and CFG. The most important thing is denoising strength. If you're going to defake in this state, you'll have to defake when your face is all changed. So, when you deake, it's better to just set the denoising strength to 0. You have to convert the face to fit the image size, so press the triangle here to check the size. Finally, click the loop, drag the face like this, and turn on the button. Now, we're all set. Now, what do we have to do? We can create it. In fact, you don't need to enter a positive or negative prompt. It's all in, but that loop is all processed. You just have to think of it as changing the face at the end. So, if you don't need to change the existing image, just set the denoising strength to 0 and put a loop on it. I'll try to create it. Create in this state. Now, the results are out. I'll take a look at the image. The left is the original image. And the right is the image of Angelina Jolie. It's a deepfake image. It's creepy. It looks like a deepfake function has been applied too naturally. It takes time for us to use Photoshop or other tools to synthesize people like this. Or you can use a paid app. If you use Stable Diffusion like this, it's amazing that you can change your face in a few seconds. It's a really amazing and scary world. The content related to deepfake is not deep. When I give a lecture on this on YouTube, I often cut it. So there are a lot of fun features, but today I'm going to use this loop. I explained that you can deepfake like this. And there's an expansion function called Reactor, not a loop. It's a deepfake expansion tool that can only change a few images to a certain person. I'll talk about that expansion function again at another time. Today, I've only covered this loop. You may have a lot of errors when installing it while using it. There may be a frustrating part. In that case, come to the comments or Naver cafe and leave it on the bulletin board. I check it right away whenever I have time. I hope you use it well. That's it for today's lecture. I'll prepare more useful content next time. Thank you.
Info
Channel: 트로메들로아
Views: 1,462
Rating: undefined out of 5
Keywords: #ai이미지프롬프트, #딥페이크보는법, #스테이블디퓨전프롬프트, #스테이블디퓨전설치, #ai이미지편집, #스테이블디퓨전학습, #딥페이크기술, #ai이미지판매, #스테이블디퓨전영상만들기, #ai이미지무료, #스테이블디퓨전컨트롤넷, #딥페이크시청, #딥페이크광고, #ai이미지변환, #스테이블디퓨전애니메이션, #스테이블디퓨전설치방법, #스테이블디퓨전노트북, #ai이미지생성, #ai이미지만들기, #스테이블디퓨전맥북, #딥페이크시청처벌, #스테이블디퓨전사용법, #ai이미지사이트, #ai이미지영상, #ai이미지저작권, #스테이블디퓨전ai, #스테이블디퓨전, #ai이미지, 딥페이크 만들기, 딥페이크 프로그램
Id: HVXEKojkuZM
Channel Id: undefined
Length: 12min 38sec (758 seconds)
Published: Tue Jan 09 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.