No updated open poses, no Hands Depth Maps. So, in this video, we're going to look at how to adjust your hands, which is a problem that many people have. The root cause is that there are many complicated configurations, so there are many things to put in, but the size is relatively small. So the solution needs to be upscaled just like the face, and additionally inpainted if necessary. Now, let's create an image and make corrections. First, let's create a portrait image in the usual way. Now that the picture has been created, let's zoom in. You can see the error in the finger part. Now let's do the first correction by upscaling. Check Hires fix for upscale. Set Upscaler to Latent. I will leave the reason for setting it to Latent in the video description. Upscale by is set to 1.5 or 2. In the picture, the size of the hand must be a certain level for smooth correction. Go down to the bottom and enter the Seed value as well. When the settings are complete, move to the top right corner and click the Generate button to generate a picture. If you're lucky, upscaling alone can make your hands and fingers look pretty. Let's zoom in and check the created picture. You can see that some of the fingers have been calibrated. But it looks like it needs correction. Now, I will change the finger to the desired shape and correct it. You'll need the help of a graphics program like Photoshop to work with inpaint. This is to create a reference image in the same form as the screen. I will guide you separately at the end of the video on how to edit a reference image with Photoshop. Click the Send to inpaint button to start inpainting. Apply a mask by painting the finger part black. In addition to the finger area, you need to paint it wide enough. Go to the bottom and activate ControlNet. Load a reference image that has been worked on in Photoshop in advance. Enable must be checked. And most importantly, you must check Guess Mode. Guess Mode will be explained separately in the video description, so please refer to the video description. The Preprocessor item selects the canny model. Select the canny model in the Model category as well. The most successful results were produced when the preprocessor and model items were applied as canny. The rest of the options below are set to their default values. Go to the upper right corner and click the Generate button. You can see how it's being corrected. Let's check the created picture by enlarging it. The finger shape has improved to some extent, but it is still a bizarre shape. In this case, it may be helpful if you upload the Weight item of ControlNet. Let's increase the weight value and create the figure again. I will raise the Control Weight item by about 1.3. Go to the upper right corner and click the Generate button again. You can see how it's being corrected. Let's zoom in on the recreated picture and check it out. It's not perfect, but it's a level that can be satisfied with some correction. Unfortunately, it would be okay to correct it using a graphic editing program such as Photoshop. So far, we have guided you on how to correct when your fingers are deformed. The batting average was the best when correcting the fingers, and we provided guidance on how to create the desired hand shape. In order to calibrate the fingers with the method described above, various types of hand images are required in advance. Images are easily available through free sites. Now, we will guide you on how to create a hand shape reference image to use in ControlNet using Photoshop. First, prepare the created image in Photoshop. Upload the prepared hand image. Adjust the Opacity value to make it transparent. Adjust size and position. Adjust as precisely as possible. Adjust the Opacity value again. After editing, paint the background black. When you're done, just save it. This completes the photoshop work. We will guide you through the precautions for finger correction work. First of all, the size of the finger must be upscaled to a certain level or more to be corrected. And when the direction and angle matching of the finger is correct, the correction can be done correctly. Since Stable Diffusion produces pictures at a lower resolution of 512 pixels, upscaling alone can improve a lot. I will leave various contents that were not introduced in the video in the video description. Thank you for watching the video so far. Also, please like it.
