IntrinsicEdit: Precise generative image manipulation in intrinsic space

Fig. 1. We propose a generative framework for diverse image-editing tasks, where precise manipulations can be performed in an intrinsic-image space and global-illumination effects are subsequently resolved automatically. Here we show a progressive transformation of an input image: ➀ We first remove the flowers and the vase from the albedo channel and then ➁ insert a new object in that channel. ➂ We replace the texture of another object before ➃ relighting the scene using a new irradiance channel. After each intrinsic-channel manipulation, we can render a physically plausible result. No single prior method can perform all these edits and provide similar levels of precision and identity preservation while delivering comparable image quality.

Generative diffusion models have advanced image editing by delivering highquality results through intuitive interfaces such as prompts, scribbles, and semantic drawing. However, these interfaces lack precise control, and associated editing methods often specialize in a single task. We introduce a versatile workflow for a range of editing tasks which operates in an intrinsic-image latent space, enabling semantic, local manipulation with pixel precision