Image to Image

Image-to-Image

Using an Initial Image (available on Control Layers) allows you to inspire the AI with an initial drawing or photo, which preserves the original image's rough structure, colors, and layout, while using AI to reimagine new content with your input prompt. You can control how closely the system should stay as it generates the image by controlling the Denoising Strength, which ranges from 0.0 (keep the original intact) to 1.0 (ignore the original completely). The default is 0.75, and values between 0.25 and 0.90 give interesting results. A decent guideline for using this is:

• .35 will produce very minor variations on the image.
• .55 will produce very moderate variations on the image.
• .75 or more will produce significant variations on the image.

Important Note

The Image to Image feature does not work well on initial images smaller than the default Width and Height of the model you are using. Make sure your image is large enough to use or use the “fit image to output size” option in order to automatically scale smaller images to the size you’re looking to generate at.

How Does It Work?

The main difference between using Image to Image and a regular Text to Image generation is the starting point. While Text to Image starts with pure noise and refines it over the requested number of steps, Image to Image skips some of these earlier steps (based on your Denoising Strength) and uses your initial image mixed with Gaussian noise as the starting image. Depending on the set strength, your image will be inserted into the sequence at the appropriate point, with the right amount of noise.

See below for an example of a normal Text to Image generation, and how the generated output changes when an Image is supplied with a .6 strength.