hello. This is PD Jo. As promised last time, today we will discuss image-to-image of stable diffusion. Image-to-image is a method of creating another image using an image. Today, we will learn about the purpose and use of image-to-image options, as well as the inpaint function. I will try to explain it in as much detail as possible. So shall we begin? First, let's look at the ability to analyze prompts in image-to-image. This is a function that analyzes prompts through images of both Interogate Clip and Deep Blue. The difference between the two is that Clip creates word-type prompts that replace sentence-type prompts. First, drag the image with the mouse, then select the method you like between clip and deep blue and click. Personally, I prefer word-type prompts, so I often use Deep Blue. It tends to get weird stuff sometimes, but it's very useful because it prompts you to scrape the details. Let's add sunglasses to the prompt and create an image. An image wearing sunglasses was created with a composition similar to the base image. Let’s take a look at the options below. In resize mode, Just Resize ignores the aspect ratio and creates the image as is. Crop and resize first creates an image while maintaining the ratio of the base image, and then This mode cuts the image to a new size. Resize and Fill creates an image while maintaining the ratio of the base image. It is created by automatically filling in the space created by the new size. Lastly, Just Resize Latent Upscale is the same as Just Resize used previously, but Upscale using latent upscaling. Personally, I recommend creating an image the same size as the original image. If you scroll down a little further, you will find Denoising Strength. Denoising strength is very important in image-to-image. Denoising strength is a setting value that is influenced by the text prompt. The lower the setting value, the less change there is in the original image. The higher the setting value, the more the text prompt is affected than the original image. There is a big change in the image. This is a step-by-step change in the setting value of each denoising strength. Please test by changing the settings little by little. If I keep the original image and make some changes, such as adding accessories. Use it by adjusting it between 0.3 and 0.5. If you want to make a lot of changes, set it to 0.5 or higher. Next is inpaint. Inpaint is a technique that uses masks to turn specific parts of an image into text prompts. Let's give it a try. First put the image. The size of the brush can be changed by clicking the pencil icon. Apply the mask to the shoulder area. Clear all prompts and just insert the tattoo. The tattoo was added quite naturally. Let’s take a look at InPaint’s options. Mask blur is an option to change the mask area smoothly. If the mask blur is low, the contrast with the surroundings will be clear, and if the mask blur is high, the image will blend well with the surroundings. There are two mask modes: Inpaint Masked and Inpaint Not Masked. Inpaint Masked changes the masked area and does not change the remaining areas. Inpaint Scythe Masked replaces all areas except the masked area. Inpaint Masked is used to remove or replace part of an image. Inpaint Not Masked is used to change the background, etc. Masked content is a setting for how to draw the masked area. Fill fills the masked area with a color that matches the pixels. Original fills the masked area similar to the original. In delayed noise, the masked area is filled with random noise, and in delayed nothing, the masked area is filled with 0. If you have a simple background and want to work with it, it's good to use fill. Original is good for preserving the original of the masked area. Latitude noise is good when AI needs to creatively fill the area. Latent Nothing is great for removing unwanted elements. This time, let's erase the cars on the road in this photo. Paint the cars in the photo with a brush. Select Latent Nothing from Masked Content. how is it? The car was simply wiped out. The inpaint feature has a lot of different uses in a lot of places. You can change your outfit like this. And the photo where the finger is shown incorrectly. You can also modify just your hands by putting a mask on them. You can also freely design the interior. I hope you all use it well and create many creative images.
