Generating digital images in new contexts while preserving color and composition using diffusion neural networks

Abstract

The present disclosure relates to systems, methods, and non-transitory computer readable media for generating digital images utilizing a diffusion neural network to preserve color harmony and image composition from a sample digital image while modifying image content. In some embodiments, the disclosed systems receive, via user input, a text prompt defining query image content and a sample digital image depicting a color harmony. In some cases, the disclosed systems generate a blurred digital image by blurring pixels of the sample digital image while preserving the color harmony. In some embodiments, the disclosed systems generate, utilizing a diffusion neural network, a modified digital image depicting the query image content having the color harmony of the sample digital image by denoising the blurred digital image toward a noise vector of the text prompt.