Image-to-image prompt engineering tips

A fascinating trend in generative AI, especially when prompting the model, is image-to-image. This covers a broad array of techniques that let you bring an image when you invoke the model. The response will then incorporate your source image into the response, letting you more concretely determine what response the model will provide. This is incredibly helpful for increasing the resolution of the images, adding a mask, or even introducing objects to then seamlessly format into the output image in any context.

These core capabilities are possible through a technique introduced in early 2022 (4), called Stochastic Differential Equations Edit (SDEdit), which uses stochastic differential equations to make image synthesis and editing a lot easier. While it sounds a bit intimidating, it’s actually very intuitive. It lets you add a source image to your pretrained diffusion model and use that base image as inspiration. How? Through iteratively...