Image-to-Image (often shortened to 画像から画像) starts the diffusion process from an existing image instead of pure noise. The model preserves the rough composition and colors of the input but applies the text prompt to transform style, subject, or details. Useful for converting a photo into a painting, refining a sketch into a finished illustration, or generating consistent variations of a starting image. Every major model supports some form of 画像から画像; the implementations and strength sliders differ.