Image-to-image
Also known as: img2img, image conditioning, style transfer
Image-to-image generation bridges the gap between editing existing content and generating something entirely new. You provide a source image and a text prompt, and the model uses the image as a structural anchor while applying the transformation described in the prompt. A rough sketch becomes a finished render. A product photo gets restyled to match a brand aesthetic. A landscape image is transformed to a different season or lighting condition.
The amount of influence the input image has on the output is typically controlled by a parameter (often called 'denoising strength' or 'image weight'). A high influence setting keeps the composition close to the original; a low setting lets the prompt dominate. Midjourney's image weight parameter (--iw) and Stable Diffusion's img2img mode are well-known implementations.
Image-to-image is also the underlying mechanism for video generation from images, virtual try-on (taking a clothing item and placing it on a person), face consistency tools, and architectural visualization workflows where a rough model or reference is converted into a photorealistic output. Most AI image generation tools now offer it as a core capability alongside pure text-to-image.