This face repair extension can make you change from this to this, can't you see it, then you should be able to see it from this to this, right? stable diffusion The face and hands have always been a bad draw when drawing a full-body image in low resolution. This face and hand repair extension can perfectly repair the face, but the hand part needs more work . Today's main introduction is stable diffusion! After Detailer extension Face repair part 1 ! After Detailer installation Go to the bottom of the video and click Installation link Click the drop-down menu , click the copy link button, go back to Stable diffusion , click extensions , click install from URL , paste the installation URL you just copied here, click the install button , there is an error because the old Abbe forgot to turn off the VPN, turn it off, turn it off After pressing the install button, this line appears on the screen, indicating that the installation is complete. Click installed , then click Apply and restart UI , and the installation is complete. Click Settings, pull down the screen to the far left, find After Detailer, click it to enter, and then pull the screen up. By the way, you can set how many After Detailer Models to have. Originally, 2 old Abe set 3 to try. This option asks if you want to save the preview view of the mask . This is whether you want to save the Adetailier processed image. The following 3 Old Abe doesn’t know its function. If you know it, please leave a message and share it. Before clicking apply settings, let’s explain this one . This one is to set the number here . Now there are two. We just set 3. Let ’s click apply settings to save Set and click the restart UI button and the number here will become 3. The second part is an introduction to the principle of face repair ! After Detailer face repair is to automate the inpaint (partial redrawing) in the stable diffusion img2img (picture generated image), so that you don't have to Paint over there and repaint the part of the painting mask. I think everyone should be familiar with it. I don’t want to waste everyone’s time here . The principle of face restoration is roughly like this. There is an original picture that needs to be repaired . Program Locate the center of the face in the original image, and then locate the mask range of the face . Increase the mask range and throw it to stable diffusin inpanit for local redrawing. Redraw the mask range, that is, the face side. What you know is face detection, mask masking and inpaint partial redrawing Part 3 ! After Detailer Simple face repair skills After you install the extension! After Detailer will appear in the Vincent map and the map . Let 's go first Go to Vincent’s picture and run a picture of the face collapse. It’s very simple to run the picture of the face collapse. Just set the width to a small size and then use the full body picture. The probability of face collapse is almost 80% to 90%. Let’s see if there is any collapse. Well ... the collapse is very bad . This is because we set the width to be small, and it is a full-body image, so there are fewer spots that can be placed on the face, causing the face to collapse. Let’s use the installation just now Yes! Check the After Detailer extension to fix it. Click to open it. Pull the picture here first . This is the detection . This is the pre-processing of the mask mask . This is the inpaint partial redrawing . These three are the three we talked about in the second part . This should be Do you know why these three are here? Next, pull up the screen to enable Adetailier. If you want to enable this function, you can tick it. You can place more than one label in this place. Different labels can be used for different masks. Processing This drop-down menu is! After Detailer Identification Module The module will be downloaded automatically when it is used for the first time . There are several modules for the face, hands and whole body . Let’s set the face first to try it out . This is The partial redrawing prompt is a partial redrawing negative prompt. In fact, its preset value has already met most of the situations . As long as you tick the enable here, you can handle most of the pictures . Let’s try it out. It’s stable in terms of Vincent’s picture. Diffusion will run a picture first, and then throw it to After Detailer for processing. Now stable diffusion is running the picture. After running this stage, After Detailer receives the picture and is doing face positioning, detection, masking and reconstruction. Paint repair, you can take a look at the position and range. The position and range can be adjusted manually . After the run, you can see the effect. The effect of the face repair is very good . Open the picture and zoom in. This is before repair and this is after repair. It’s really not easy to do this with such a low resolution . Let’s take a look at the original image again. It’s really quite different . Back to stable diffusion, let’s talk about the three settings. The first one is detection and this one is Detection model confidence threshold is the detection
rate. The higher the value , the worse the detection effect. Even a face can’t be detected . The lower the value, it means the detection ability is super strong. Not a face, sometimes it will be detected as a face. It ’s normal . Under normal circumstances, a single person can be set to about 80% of the faces of multiple people. To detect the faces of multiple people, the value must be lowered, and the value should be pulled up to 100% for everyone to test once . In theory, only very large faces can be detected. This kind of small face is usually not detected . Normally , after the stable diffusion runs, if the face is detected in After Detailer, it will show the area to be masked . There is no indication here! After Detailer did not detect the face. Therefore, there is no face repair process . If you have such a situation, just adjust the place just now . Set it to 80 and try it. After setting it to 80%, you can detect the face and enter the mask pre-processing and face. Repair The function of this slot is probably like this. OK, the face repair is completed . Next, there are Mask min area ratio and mask maximum area ratio. These two are when the mask area is too large, it will be modified to non-face The internal part , if this is the case, then come here to adjust the minimum and maximum by dragging and dropping. Generally, it is recommended to put it in the preset. The detection part is probably like this . Then there is Mask Preprocessing. Click to open these three options. The control is just the stable diffusion run, there will be a zoomed-in picture. This is the one. Grab the previous video and take a look . This is the one. You can manually control this range . Let’s try it. The first one is to control the x-axis , and the middle one is to control the y-axis. The bottom one is the size of the control range . Some people may not be able to understand it. Let’s see it more clearly . First, reset the x-axis and y-axis to zero, adjust the size of the range from 32 to 12 , and run the graph to see this change. Just change the range from 32 to 12, so the frame of the range will shrink. Which frame is that frame ? It is this frame . Have you noticed that the range of the frame has become smaller? This is the frame of the previous range of 32. Can you see the difference? The range Manual adjustment can allow you to mask more details, but there is not much difference in low resolution. Ok, let’s take a look . In fact, it’s really hard to see the difference in this low resolution. Next is the x-axis and y-axis. This is easy to understand. It is the box. If it is not in the correct position, you can manually move it around. Setting the Y axis to +60 means moving up 60 units. The old Abe didn’t go into it. It should be pixels. Let’s see how the box changes . Do you still remember the box with the y-axis at 0? Did it move up 60 units , so that everyone should be able to understand the role of those three parameters . Let’s take a look at the effect. Why is there no fix? Because the frame is pulled up too much, the face is not in the frame, so there is no way to make a mask. First change this back , and then the following three. This should be related to whether there is a new mask tag above. You should be able to understand it by reading the text. The function here is not to waste everyone’s time to explain more . The first detection is finished, the second mask pre-processing is also finished , and the third inpainting partial redrawing is the same as the inpainting partial redrawing of stable diffusion. Take a look. Click here . Its settings are the following settings. I think everyone is familiar with it. I don’t waste time explaining it here. Those who are not familiar with it will give old Abe a bowl, and I will share the video when I have time . The last one This controlnet model only partially redraws the stable diffusion inpaint and replaces it with the inpaint of controlnet. Here is the weight adjustment of controlnet inpaint . Then , the operation method of the picture generation picture is the same as here. If the video is helpful to you , please remember to subscribe, share and press the small bell for the old Abe . Thank you, everyone. I wish you health and happiness.
