【Stable Diffusion】既存の写真から好きな画像を生成できる「img2img」を使って人物、景色、建物等をカスタマイズ！！

This section introduces the topic and purpose of the video. The speaker, KJ, explains how to use Stable Diffusion to generate images of natural landscapes, animated characters, and architecture. They will demonstrate how changes to the denoising strength (from 0.2 to 1) affect how closely the generated images align with the original prompt.

In this part, the speaker guides viewers on how to generate images of natural landscapes. They start by setting the denoising strength to 0.2, which produces images closely resembling the original ones. As the denoising strength is increased to 0.5 and 0.7, the generated images begin to follow the prompts more, moving further from the originals, such as adding a lake and mountain to the scene. At a denoising strength of 1, the images completely diverge from the originals, with unexpected elements like an ocean appearing.

This section focuses on generating architectural images. With the denoising strength set to 0.2, minor changes occur, such as the hotel name disappearing due to the negative prompt text. At 0.5, the image shows a signboard that wasn’t part of the original, still maintaining much of the original design. When the strength is raised to 0.7, the architecture becomes more abstract, resembling something designed by the speaker, with more changes. Finally, at a strength of 1, the original design is almost entirely replaced with a luxurious golden building.

In this part, the focus shifts to generating images of people. At a denoising strength of 0.2, there are hardly any changes to the original. When set to 0.5, there are slight atmospheric changes. By increasing the denoising strength to 0.7, the image becomes more dramatic, even resembling a murder scene. At the highest strength of 1, the character’s face transforms into an Asian-looking person.

This section discusses generating images of animated characters. At a denoising strength of 0.2, there is minimal change to the original image. When the strength is increased to 0.5, there are subtle stylistic differences. With the strength set to 0.7, the character’s appearance drastically changes, and they now hold a camera. Finally, at a strength of 1, the image becomes entirely different and unrecognizable.

In this conclusion, the speaker shares their experience after generating many images, suggesting that a denoising strength between 0.7 and 0.86 is ideal for creating high-quality images. They encourage experimentation with settings to achieve the best results and wrap up by asking viewers to like and subscribe if they found the video helpful.