스테이블 디퓨전 컨트롤넷 ip2p 간단한 사용법 | Stable Diffusion InstructP2P tutorial

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
Hello, this time in Stable Diffusion One of the control net functions I will explain the function called Instruct P2P. This function is an image-to-image or in-paint It's one of the control net functions that can easily give you an effect in text-to-image. If you don't know how to use it, you can just pass it without thinking. If you know how to use it, it's a very useful extension Try using it with me this time I hope you apply it to your work once. To use Instruct P2P First, a control net extension function must be installed. For those who have not installed the control net yet I'll put a link in the description section If you haven't installed it yet, install it You can go back to this video And when using Instruct P2P All functions of the control net need a model So if you haven't downloaded the model yet I'll make it a little easier for you to download the model I said I opened a homepage It's a homepage called SD Hotline You can easily download control net models Actually, I downloaded it , but I was so lazy that I made it myself. If you look here, IP2P is Instruct P2P If you click this model download, you can connect to this download link and get it on your PC. Where do you put the downloaded model file? Let me briefly explain here I installed it on C drive So come into the stable diffusion web UI folder You can go to the extensions folder Click and then come to the SD web UI control net folder Finally, come to the models folder Here you can put all the control net models you downloaded So you can put the Instruct P2P model file you will use today here Now then you are ready to use the stable diffusion IP2P Instruct P2P Now, run the stable diffusion and come to this screen Now, you don' t have to go into image to image here I'm going to work on this text to image You can check the quite amazing effect through IP2P Instruct P2P The preparation is one photo First, let's go down to the control net Go down and open the control net tab like this I'll drag a pre-prepared image here And the control type is what I called Instruct P2P today Click Instruct P2P If your model is properly downloaded, it will be automatically selected If it is IP2P, FP16, it is ready But the preprocessor is set to non You can say it's weird, but the preprocessor is not applied here So you can see that it is a meaningless function even if you preview it So this image in this control net The structure of a new image created by combining this text You can know it as Instruct P2P Now, let's create it I'll show you how to create it To match the size of the control net image Press the arrow next to this camera It matches the size of this image If you press it, you can see that the size of this image is currently 512 x 510 Then I'll set the sampling method to SDK I'll set the sampling step to 30 But the result will come out a little later When I do this, the result won't come out properly I'll show you Now, bring the negative as it always applies This man looks so cold I'll put in the heat I'll put in the words like fire, flame, simply And let's see how this image changes Image creation, click Now the result is out I'll check it out This image is the first image that says I referred to this image in IP2P control net How did this image with fire and flame come out? It turns out that an image with no relationship was created The image I want is the image of this man I wanted to make an image of a man who was covered in flames I'm not wearing any clothes with a completely different man It's just a man covered in flames Now, there's a problem Usually, I give up in this state What is this? It's a meaningless control net function There are quite a few people who don't use it I was like that at first I thought it was the most useless function I applied it again later It's one of the really good control net functions I changed my mind again Now, why did this image change to another image without consistency? And secondly, how can I keep the consistency of the image I uploaded? I'll explain if I can add new flames or flames The method is quite simple If you look below, there is an answer If you look below, control weight and starting control step, ending control step This is really important Starting control step and ending control step You can just leave the default value as it is But the most important thing is control weight The fact that the control weight is 1 now It means to keep the frame at an ordinary level So, if you give control weight 1 in this man's image If you use Instruct P2P It means that the frame can change to some extent 1 is also a slightly strong control weight You can see that it has changed a lot at the level of 1 Now, you can see that the frame is too weak with this control weight 1 So, I'm going to make this control weight completely powerful, almost like a concrete frame It's simple, you can give the value as 2 You can give 1 to 2 You can give it from 0.1 to 2 Control weight So the maximum value is 2 I'll give it 2 and make it Then, leave this frame as it is An image with this flame can be created Let's create an image Generate, click The image is created Now, looking at it like this, the The original frame remains the same You can see that it has changed to an image of a flame But you have to think carefully about the image quality The size of the image can also affect the quality of the image 512x512 size is a very small image size You can't make a high-quality image I recommend that you consider that and make it Now, if you look at it, it gives you a control weight of 2 Now, you can see that it is well represented by the text flame that I wanted without changing in this frame Now, here you can say that the frame can change a little I want to make a little more change to my outfit or face Then what should I do? After coming back to the control net I'm lowering the control weight little by little If E means to create an image with this frame perfectly If you lower it a little bit from E, the frame will collapse little by little So I'll leave it at 1.7 and create it again Create Now the result is out again Now, you can see that the control weight is a little more natural than when you gave it to E You can see that it expresses the man who is covered in flames and flames Now, there's another thing you're curious about here There's a part I didn't explain at the beginning Starting control step and ending control step Now, I'll briefly explain this and finish today's lecture Now, starting control step and ending control step It's good to understand like this Now, here from 0 to 1 You can think of this as having steps from 0 to 10 From 0 to 10 Now, but here I'm going to leave the starting control step at 0 Leave the ending control step at 0.5 This means that this control net is applied only from 0 to 5 So from 0 to 5, this control weight is maintaining this frame at 1.7 You can think of it as maintaining it and then the control net is released Now, then you can create a more free-spirited image I'll make it once Now, control net weight 1.7 Starting control step 0 Ending control step 0.5 I'll create this Click Now, the result is out Let's take a look Now, the man who was caught in the fire and this pose are the same You can see that the face has changed as a result So, the control net is at level 5 From 0 to 5, the control weight is holding the frame tightly Then it was released And then I ignored the existing frame It would be easy to think that the face has changed Now, if you use stable diffusion like this I don't know a little bit, but can I touch this? There are buttons that I want to The button that causes an error in the system is not necessary in this place Don't worry about it If you don't know, you can touch it once Checking the results can also be a good way to improve your skills Now, today, we learned about the control net's Instruct P2P function Even without going through image to image In the text to image, with this control net function I put a variety of effects in the picture We learned an amazing function that can be created If you use this function, you don't need to use any other tools You can easily put the effect you want I recommend you to apply it to the work you want That's it for today's lecture Next time, we will prepare more useful content and meet you Thank you Thank you for watching
Info
Channel: 트로메들로아
Views: 333
Rating: undefined out of 5
Keywords: stable diffusion, 스테이블 디퓨전, ip2p, Instruct2p, 스테이블디퓨전 컨트롤넷, 컨트롤넷, controlnet, how to use ip2p, instructpix2pix, ai, aiart, ai explain, ai 이미지, ai 그림
Id: JRTP9hYoST0
Channel Id: undefined
Length: 11min 15sec (675 seconds)
Published: Wed Feb 21 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.