Image to image (img2img) with Stable Diffusion
Table of contents
Use image-to-image to take the features and structure of a starting image and reimagine them with a prompt. The colors in your original image will be preserved.
Prompt strength (or denoising strength)
This only applies to image-to-image and inpainting generations.
It determines how much of your original image will be changed to match the given prompt.
Higher numbers change more of the image, lower numbers keep the original image intact. Values between 0.5 and 0.75 give a good balance.
When inpainting, setting the prompt strength to 1 will create a completely new output in the inpainted area.
Using prompt strength 0.95
In this example we’ve increased the prompt strength. You can see that the image has changed a lot, it matches the prompt more closely and our original cat is very much a tiger.
Using prompt strength 0.45
If we do the opposite, and reduce the prompt strength, you can see that much more of the original image is preserved. Our cat has only slightly changed, and it’s showing minor tiger-like features.
