Abstract
Recent advances in generative imagery have brought forth outpainting andinpainting models that can produce high-quality, plausible image content inunknown regions. However, the content these models hallucinate is necessarilyinauthentic, since they are unaware of the true scene. In this work, we proposeRealFill, a novel generative approach for image completion that fills inmissing regions of an image with the content that should have been there.RealFill is a generative inpainting model that is personalized using only a fewreference images of a scene. These reference images do not have to be alignedwith the target image, and can be taken with drastically varying viewpoints,lighting conditions, camera apertures, or image styles. Once personalized,RealFill is able to complete a target image with visually compelling contentsthat are faithful to the original scene. We evaluate RealFill on a new imagecompletion benchmark that covers a set of diverse and challenging scenarios,and find that it outperforms existing approaches by a large margin. Projectpage: https://realfill.github.io