UniTune: Text-Driven Image Editing by Fine Tuning a Diffusion Model on a Single Image

Text-driven image generation methods have shown impressive results recently, allowing casual users to generate high quality images by providing textual descriptions. However, similar capabilities for editing existing are still out of reach. usually need edit masks, struggle with edits that require significant visual changes and cannot easily keep specific details the edited portion. In this paper we make observation image-generation models can be converted image-editing simply fine-tuning them on a single image. We also show initializing stochastic sampler noised version base before sampling interpolating relevant from after further increase operation. Combining these observations, propose UniTune, novel method. UniTune gets as input an arbitrary description, carries while maintaining fidelity does not additional inputs, like masks or sketches, perform multiple same without retraining. test our method using Imagen model in range different use cases. demonstrate it is broadly applicable surprisingly wide expressive operations, including those requiring were previously impossible.