Hello, thank you for your hard work getting here.
Wasn't it hard? Today is the last optimal setting STAGE. I'll be there right away Today, I'm going to briefly explain the image-to-image screen. I'll focus on inpainting and outpainting. If you look at it, I will try to create an image once. Now, I will take a look at this now. When I make a picture, I always look at my hands first. In general, I think the control net or such inpaint functions are better in animation than in live action. Then, I will transfer this as an image to image here. This information will then be automatically transmitted. Then here's a function like this If you want to change the image back to a real image, you can change the model once. For example, let's change this real-life model. It took some time to change the model and recognize your model. Let's create it as it is without changing the settings. Now, if you want to fix this seed, you can get the seed fixed number from text 2 image, copy the number here, and put it here now. I'll try to create it like this. Yes, it was created right away. Because I changed it in 2.5D, the face is not perfect. Yes, I think you just need to know a little bit about this concept. Conversely, you can also try using the function that can change live action into animation like this. Yes, then I will look into the sketch function next time Yes, I've been talking for a while, but I can't record, so I'll do it again Now, I'll put a picture I drew on top. Yes, I wonder how it will come out with this sketched look. Of course, I have to change the prompt now, but it looks like this with my face Who do you think you look like? Anyway, your image is being created. This pretty image was created. It is a function that allows you to create an image based on this sketch to some extent. If the denoising intensity is reduced a little here, you can create another image. Let's also reduce the CFG scale value and set the denoising value to 0.4. Since it came out only with a focus on sketches, yes, you can see that the picture was created with a slightly different appearance from the picture by changing the denoising strength and scale a little, properly mixed with the prompt. Yes, a new image came out like this, right? Yes, I think I can have fun with this. Next time, let's look at the inpaint function that we need to focus on. Now then, I'll try with the Wonder Woman picture from earlier. Now, take this and send it to In-Paint. will work here again Now, what to do here is, if you enter the inpaint function, you will have a tool that can paint like this. But with this, I can now adjust the thickness. So here I want to put a mask on this Wonder Woman for example Then make it in the shape of a mask like this. It's good to do it precisely, but you can roughly paint it like this Nothing much has changed here and at the bottom. I'm going to process only the mask, so select it as it is, and this part is a bit confusing, but there is a marked process like this. If you fill in the marked place, it means that the part inside it is filled with another part. Now, keeping the original means that you change the elements in this while keeping it as it is to some extent. I don't know what you mean I will show you this difference in two steps. Now, in the Inpaint area, only the masked part will be changed. Yes, we will proceed like that. Then I'll run it here and adjust the adjustment part again. I don't think this is particularly necessary here. For example, I'll try mask. I'll look at the mask and create it. Yes, then the image was created like this, but it was made for fun. This is the image that can come out when you select Keep original. The face is not completely covered. To some extent, the image was completed with the original image and the surrounding color, but if you fill it in, a mask will be created. For example, let's change it to our golden mask. As it suits Wonder Woman, yes, if you check it, the mask is now warmed up like this. I think it will be easier to understand if you know the difference like this. Yes, the inpaint masking function is also used in hand correction, so I think you can use this function. Later, I will make another production related to hand correction. Yes, and then I will do an Inpaint Sketch. Inpaint Sketch is a combination of the Inpaint function and the sketch function. yeah i'm pulling it And I'm going to copy and paste this phrase again here with this original prompt. yes in this image i will now paint Because it is a function that combines sketching and in-painting, there is a brush and you can select colors. You can also choose the color I'll adjust the brush size a bit and try it on the whole face. This function is actually not a part that I use very usefully, so I'm going to show you something once, so I'll try to create it like this. We need to make a little more adjustment here. I will keep the original Because this is selected as the fill right now, let's change it here and create it once. Yes, if you look at it like this, you can now see that the picture and the image of this certain prompt have been synthesized. Yes, in a way, it's deepfake technology, but it's very awkward. If you want to soften this boundary value a little bit, you can increase the mask blur to some extent. If you create it in this way, I think you can see a little bit of harmony together. The borderline is so clear yes it is You can see that the edges are slightly smoothed to some extent. Looking at it over and over again, it's not really strange, but I think it's because there's a picture and a real image like this. great And now, I think I need to interpret the concept of inpainting, filling, and maintaining the original a little differently in this part. Keep original means to keep the image of the prompt, I think you can interpret it. Alright and let's move on Uploading InPaint will bring up two windows: It is to implement a new image by taking the original image and the masked image. Some people may need this function again. Now, I'm going to copy the Wonder Woman image here as it is. Okay, and with this, I'm going to crop some of this area in Photoshop now. I imported the image into Photoshop. If I click the quick selection tool in the left pane and make a selection like this, the image is automatically selected. yes it happened It's very convenient these days, so I was able to select an image right away. Yes, I can confirm that even the hole here has been excluded. Roughly this is what I get Then press ctrl+J to copy the image layer. And if you press the control button and press the image, you can see that the area you selected earlier is selected on this layer. Now, I'm going to paint black here. If you press Alt+delete, the foreground color (black) will be painted like this. Yes, I'll change the background to white here too. Select a new layer, and I think I said it wrong. To create a new layer, it's Control+J, not Control+z. If you do this, it will be a layer copy. Now, in this state, I will sprinkle white here. If you do ctrl + delete, the back color (white) will be filled like this. I'll merge this into layers (using ctrl+E), merge four layers, and save. Yes, then I came to the inpaint upload window. I will put pictures in each of these two sections. I'll put the original, and then I'll put the file that I worked on with Photoshop like this. Then I want to change the background, but I'm going to delete all of this and let it be in the forest. In this way, let's just leave the following conditions as they are and change them. If you look at the forest... It's not very elaborate, but you can see that the background is filled with images in the forest. In this way, you can think of it as a function that can be used for background synthesis or partial synthesis. Now, if you look at it, you can see that only the mask has been processed. In fact, in the inpaint function, black was the mask area. It can be a bit confusing, but in the inpaint upload window, you can see that the white color has become a mask. You may be a little confused because of this part, but I think you need to distinguish the two parts. Yes, and I changed it from the original state, and if I fill it, it seems that another image will be created. And since it is a masked part, you can see that the white part has changed. You can do this Let's try filling I think this image came out a bit more natural when I chose to fill it with four. In the case of the background, this is more natural than the previous image. Maybe now, if you keep the original, you can do this because you combine it with the existing information. If you fill it, it seems that this masked part is newly created to create an image. Yes, anyway, I was in a hurry and tried to tell you a lot, so time passed again. So now we've come this far and now there's the deployment part The placement part is not a necessary part now, but it is a useful function when making a video now, so let's take a look at it then. Now, let's look at the last remaining outpaint function. ok let's go to img2img first When you go, if you look at the script, there are two items for out painting. the truth is hiding here Therefore, you can choose one of these two without looking for it in the above items, and I will choose the first one. You can select this, but even if this option inside seems to be a difficult part, there is nothing to change significantly. For example, the function of out painting is this image. It means that in the state where this image exists, more images of this margin are drawn. It would mean that AI draws creatively. Yes, and in the same case as before, this live-action model was selected and the Camellia model that originally drew this model was set again and created. Alright, nothing much to change Here, in order to make the picture on the right more creative, it is good to increase the sampling step little by little. If you don't increase the number of sampling steps, the image will come out a little monotonous and strange images will come out, so increase it like this and learn the outpainting technique... The image came out like this, but it's not natural. It seems that there are AI memories of what we did here earlier. So I wonder if it would be better to delete all the work I've been working on. I don't know if it has any influence, but anyway, I'll make it possible to create it again in this state. Yes, it seems that the AI keeps remembering the existing information, so even if it is regenerated, the same result came out, so I ended the program again and made it a horizontal picture like this. But with this, I will explain the out painting function once again. Now send this as an image to image If you send me out, you will come out here alive. I'll make a choice here Choose between Outpainting Mark Two or Poor Outpainting. Here, you can see the number of pixels to expand. If you widen this, it will come out wider. It is a blur function to call a mask, and the part explained earlier and the lower the tax reduction index is, the higher the detail is. For now, I will proceed without changing this state. But here, I will do the right part as it is selected and proceed without any further changes. Yes, and I will also raise the sampling steps. I finally cleared my head and the AI created it properly Creatively made, very pretty Yes, this background was connected very naturally without any steps Let's expand the left side as well. The left side is now How to widen it, you can choose the left side. Select the left side and now create it as it is, but the thing to note here is that I wanted to widen the left side with this image. I will do it one more time, not with this image. Then this image is transferred back to image-to-image. I'm going to breathe for a while Then this image has now been moved here Yes, I will extend the out painting in this state. I will press create Your image came out. What do you think? The image is quite pretty. You can expand the background a bit like this You can expand infinitely If you continue to work, the background will come alive instead of the image getting smaller, and nevertheless, the quality of the image will not decrease and it is a very useful out-painting technique that can improve the completeness of the picture. But it would be nice if you use this a lot. It would be nice to practice this kind of image extension. In this way, we basically checked the functions related to stable diffusion. Yes, so far, I've been making things in a hurry for each series. I thought that this level would help a lot in terms of understanding, so I thought I'd finish the series here. I'm going to practice learning stable diffusion as I go. This time I did dark mode like this I'm going to proceed like this now, thinking that it would be more comfortable to see when using the dark theme. In fact, today's function contains a lot of content and is implied, so it's also a difficult content. This is not an easy part if you do not practice directly once, so if you watch the video several times and learn it, it will help you a lot and you will enjoy stable diffusion. Yes then, if today's video was helpful, please like and subscribe. I'll see you with better information. Have a nice day and thank you
