Img2img (image-to-image) is a way of using an image model where generation starts from an existing picture instead of pure noise. The model partially “noises” your image and then redraws it following a prompt, so the result keeps some of the original.
Denoising strength
The key setting is denoising strength (sometimes called “strength” or “creativity”), usually between 0 and 1:
- Low (around 0.2–0.4): the image stays very close to the original; colours and small details change.
- Medium (around 0.5–0.6): materials and furniture can change while the overall layout mostly holds.
- High (0.7 and above): big changes in style, but walls, windows and perspective start to drift.
The trade-off for rooms
With img2img alone, there is a tension: low strength barely redesigns the room, high strength redesigns it but may move the window. That is why room tools combine img2img with structural guidance such as ControlNet, and fix details with inpainting.
Compare with text-to-image, which starts from nothing but a description.