Hello, today I think this will be good information for those who want to draw pictures that bring out colorful backgrounds and characters well. If you zoom in, you can usually see that the face is blurred when the person is small. Now, even if you zoom in like this, you can see that the outline of the face is still alive to some extent. Today, I will introduce you to BMAB, which provides functions related to this. Also, there are many other additional functions that BMAB has that are useful, so I will explain them as well. I'll go quickly If you look at the web UI, you can see that BMAB is installed. This can now be installed through an extension program. So let me take a look here. This program called BMAB was created by a Korean. I think Portu is a genius. He created it very well and there are all the related explanations here. If there is something I couldn't explain here, you can check it through this part, and in today's case, I will try to explain it in chunks in a simpler way. Enter the code here, click copy, enter stable diffusion, and now enter the extension function. Go to install from url here, enter this as is, and then install. Once you install it, it will be listed here like this. You can now re-run this through apply and restart UI. I haven't checked everything, but it seems to be available starting from WebUI version 1.6. And as you can see here, these conditions must be met for it to work, so make sure to check. And it works like this in Windows, Linux, and the local environment. I will also show you the versions I used here, so please refer to them. Regarding BMAB, a subscriber asked me to cover it some time ago, but it was a bit too late. I'll explain it to you today. Here now the checkpoint model uses this I received this from here and used it. It doesn't matter which one you use. I would like to add here that depending on the checkpoint model, wide shots are good or hand and foot expressions are different, so please keep that in mind. Today, we will start with a simple prompt. It's simple to write like this, and we'll set up some background settings like this, and then we'll use this for Negative. There's no big difference. In this situation, in this case, go up to DPM and select DPM++ 2M Carras. Would you like to set this to 30? Do 30 and then High -Res like this Let's set the upscale model to 4X Ultra Sharp, set it to 24, and set Denoising Strength to 0.3. Because we want to upscale by 2x, we'll change this to 960 (half of 1920) and make it 1920X1080. So let's do this and then this will now be 540 (half of 1080), right? Yes, this is how we proceed. After that, there is nothing to change much about the rest. Here, let's draw a picture to create a wide shot without using Adetailer or BMAB. Would it look better if I added a wide shot? Let’s create it like this Yes, it was made. It's also a wide shot, but it's a bit ruined because I didn't use the face detailer. I want to make it wider In order to do that, we have to put it together like this one more time. I'll give it another weight. I'll try to create it again like this at about 1.5. As expected, it works well. It was made a little wider. Now, the face is ruined like this. In this case, it's good because the background comes alive, but sometimes the character is ruined. Then you can now use AD Taylor again. Then, I'll do this in AD Taylor so that a Korean face can appear, so that the idol's face can appear. Let's create it again like this. Yes, it came out, but since I didn't fix the seed, a slightly reduced (different scene) wide screen appeared. In this case, let's add the word extremely. In this way, I will emphasize the wide shot more dramatically. If it doesn't come out, try again by fixing the seed. I'll try it one more time (I'm trying hard to change the settings... It's my first time using the checkpoint model and I'm lost haha) No matter how hard I try, this wide shot!! It appeared a little bit by chance earlier, but it's not easy to make. Since the character is emphasized so much, it's not easy to bring out the background. Scenarios for various words, or raising the weight to 2, etc. Nowadays, it is not easy to create such wide shots with a full body. That's why you need this BMAB, which I'll explain right away. There is also some A detailer function here, so I will turn it off and use it. Turn it off so that I can use only this. I'll show you one more time here. I've changed the prompt a little (different from before). Since there was no wide shot before, please use this as a reference. Next, we will use this in this state. To activate this overall, you can check Enable. There are various functions here. In the case of Basic, it is in charge of functions that allow post-editing of the picture, such as contrast, brightness control, sharpness, and partial color that can be adjusted for sharpness, or color temperature. You can adjust the warm and cold levels. And usually, in the case of our photos, there is some noise in the case of film photos. If you upload it like this, you can add a slightly photo-like feel. You can see that you can also use the image post-editing function in Photoshop like this. Next, in the case of Edge Enhancement, if you check this to emphasize the outline, it will come out a little darker.
(The detail of the overall image increases.) Therefore, in cases like this animation, it will be a more useful function. If the image is blurry, you should definitely use this. In the case of imaging, it is a function that can be used in img2img. In the case of person, you can think of it as a function that identifies the person and inpaints the person again in detail. If you look here, there are a lot of option values. But there is a reason why it was decided like this Since this is the most optimized value, there is no reason to change it significantly. If something is wrong in the picture, you can correct it after looking at the results. Now, just check this and activate it. Next, this second feature is an experimental feature. I don't think it really mattered whether I checked it or not.
(It uses GPU, but I haven't tested it much) For now, you can check it first and use it. I will briefly mention the remaining functions. Now, in this case, it functions well. If the image area exceeds a certain standard, there is no need to redraw it, so you can activate this function. Then, when a person is found, it is now called an upscale ratio, and the function increases it by 4 times and inpaints it. no see Therefore, even if the character appears small compared to the background, you can think of it as a feature that allows for a more detailed depiction of the character because of this function. Next, in the case of Denoising Strength, if the person appears large, you can increase this value. In the case of the Dilation Mask number, you can think of it as a value that sets the mask of the person found. Next, in the case of CFG Scale, you can think of it as a concept that we all commonly know. Then, in the case of the now important Large Person Area Limit, it states that if the area occupied in the image exceeds this value, it will not be redrawn. It seems to be a limited situation. There is no major problem if you use the default value. Then, in the case of limit, there are times when there are multiple people in the image. In that case, if there are too many people, start inpainting from the largest area unit and stop inpainting after that. So, if we say there is only one central character, we can set it to 1 like this: If there are 2 people, you can do this with 2. In the case of Face, it is responsible for a similar function. To do face detail, we turned off A detailer, so please check this. Next, this function now performs facial correction before HiRes Fix. So, it would be good if you check this. Then, in cases like this, if there are multiple characters, you decide in what order you want to inpaint them, but usually you just use the score. In a case like this, if there are 2 people, 3 people, or 4 people, the number will change. If there is 1 person, just set it to 1. If there are 3 people, you can write a prompt for each of the 3 people like this. Next is the part where you can adjust detailed parameters. There is a document I gave you about this earlier. I think a clearer explanation will be possible if you refer to it there. Then, in the case of the grounding dino here, it is a detection function that can be activated in the existing DDSD or the current BMAB, and is said to search more broadly. In the case of the old D detailer or the current A detailer, this model is used. You can choose like this. You can do this, and I don't think there's much of a difference.
(There may be differences between actual and semi-real images.) Now, set it like this. Usually, you can set it to this default value. In the case of a sampler, you can select the model you used initially. Let's try running it like this: I would like to paint things like hands more clearly. In this case, you can activate this function. Subframe or each hand, inpaint each hand There are these option values. Please refer to the document earlier for a detailed explanation of this part. They say it is better not to write a prompt when selecting a subframe. And in the rest of the cases, there is a recommended value to set the denoising strength according to the value of this option. If subframe is selected, set it to 0.4, and there is no reason to change the rest much. If I explain it one by one, it would be too long and a bit complicated. If you don't want to correct your hands right now, you can uncheck this. Then, in the case of resizing, I'll explain it later in Image 2. In this case, I'm not sure because I'm still experimenting. Next, in the case of Config, it seems to be a preset used here, but I don't know exactly. Now, let's do this so that we can run it using only what we have checked. Next, when it is set to 1 in Basic, you can see that there is no significant change because it is the default value. Now, let's create it like this. Now, since this is an AI-generated image, the correct image did not come out at once, so I changed the prompt briefly in the middle. In this way, I emphasized the scene, emphasized the wide shot, and then adjusted some weights and such, and the image came out like this. As I look at it, what I wanted is starting to come out right now. The background is coming out a lot, and the character is somewhat small, but now it comes out full-bodied like this. Because we didn't retouch it, it doesn't come out more clearly, but if you look at the atmosphere of the image, you can see that we've chosen the image we wanted. However, as I mentioned earlier, this checkpoint itself is not a model that makes much of the background. It usually comes out well, but when I filmed today, it didn't come out very well. So, I will make some changes to the checkpoint model so that I can write it. I used a different model with the same prompt. It doesn't matter which model it is. Try it and if it doesn't work, try a different model. Live-action models tend to come out well. In this way, I was able to utilize the BMAB function, which allows the background to be vivid and the person to appear clearly in the background. I don't know if it's called Bmap. The purpose of the creator of this BMAB function was to bring out the details of people in the landscape, so we checked the main purpose of the BMAB function, and now there is more to it than just this. There is also another useful feature that can save several people from the pace I mentioned. Rather than creating this part directly, I will show you through an example. I created three people like this and selected them without using any detailing function. Then the face came out a little strange. Then, in BMAB, I set the limit to 2 in the face function tab. Then, I selected the score, which means that corrections will now be made starting from those with larger areas. Then, since the correction is done starting from the largest area of the face, only two people were corrected and the results were provided. So, if you want to play with three people, you can set this limit value to 3. Let me tell you about the face part that can be used when multiple people show up. Next, we will check this image and resize function in Image to Image and finalize it. (Let's open our eyes and go :::) I'll go image to image. Now, I'll bring you a photo. I'm going to import a photo with a slightly larger proportion of the face. This is it. In this case, originally, if you send text to image, this phrase would appear, but I don't think it appears now because there's an error. So, I bring a positive report here. I will write this down, and then I will bring the correction report and write it down. I will proceed with this written down. Now, if you look at it here, there is nothing to change significantly, and the sampling method needs to be changed. Change the Sampling Method to DPM++ 2M SDE karras. There is nothing to change in particular. Since the size is 1024, let's go as is. Here, I will reduce the Denoising Strength a little more. Change it to about 0.15. Next time, leave the seed value fixed as is and then turn on the BMAB function. Here, check the enabler. What we're going to do here is the Imaging function. Click on this. We're going to bring in a background here. I'm not sure if this suits me, but let's put the image here. To activate this, you need to check blend enabled, and you need to check this as well. You can pull up the shape of this woman in this image, but the prompt here says one girl. So, when you type 1 girl, we automatically select this key. What I want to do is use this function to insert a random image into the background. To do that, change this number to 0. In this state, I will now return it and Generate it. Yes, the image came out like this. If you look at it, the laying down was not done perfectly. Maybe it's because the dark part here is too blurred. Anyway, when the value is 0, the character lives well and is in the background. If you create this at 0.5, two images will appear, creating a transparent image that looks like something is overlapping. If you set it to 1, an image will be created with no people at all and only this background. So I think these two images are blended. You can do this, and you can automatically use this detection function to put one person on a different background. In fact, if you are not standing right in front of the camera, the person and the background do not look very natural. If you want to make the person a little smaller, use the Resize function. Now, check here and lower this as much as possible. give If you lower it, the person will appear a little smaller. And there is this mode. There are three modes. Among these, the creator says that the InPaint function is the most useful, so you can select it like this. I think you can proceed with the remaining numbers without much change here. And there is a part that I failed to mention in Resize earlier, and the Upscale function is included here. So, for example, if you set the upscale ratio here to 2, the image will be twice as large. If you use this text 2 image with Hire.Fix, the image will be 4 times larger because 2 times 2. If you want to use this, check here and you can make the image larger. For now, let's uncheck it and create it. An image has appeared, and compared to before, the proportion of people has decreased. Since this image was also used as an outpainting function, you can see that images that were not there before have started to appear like this. If this image had been used without imaging, the background would have been preserved and only the person would have been rendered small. In this way, you can use the resize function when adjusting the proportions of a person within one frame. There are a lot of additional features like this. As we saw earlier, the main point now is that we absolutely need this BMAB for wide shots and landscape shots, and even though I omitted a lot in the explanation earlier, a lot of time has passed. So there were a lot of missing parts, and I was also concerned that I might harm the creator's intention by explaining things incorrectly, so I did that to avoid giving as detailed an explanation as possible. Instead, in the part I am linking to, there is a friendly explanation from the creator, so I think it would be good to refer to it and use it as a good feature. Okay then, we will end here. Thank you for watching today as well. Please like and subscribe. I will come back with good news next time. Thank you.