Diffusion Self-Guidance for Controllable Image Generation

TL;DR

This paper introduces self-guidance, a novel method that enables precise control over diffusion-based image generation by leveraging internal model representations, allowing complex image manipulations without additional training.

Contribution

The authors propose a self-guidance technique that guides diffusion models using their internal signals, eliminating the need for extra models or training for controllable image synthesis.

Findings

Enables control over shape, location, and appearance of objects.

Allows complex image manipulations like merging and editing.

Works without additional training or external classifiers.

Abstract

Large-scale generative models are capable of producing high-quality images from detailed text descriptions. However, many aspects of an image are difficult or impossible to convey through text. We introduce self-guidance, a method that provides greater control over generated images by guiding the internal representations of diffusion models. We demonstrate that properties such as the shape, location, and appearance of objects can be extracted from these representations and used to steer sampling. Self-guidance works similarly to classifier guidance, but uses signals present in the pretrained model itself, requiring no additional models or training. We show how a simple set of properties can be composed to perform challenging image manipulations, such as modifying the position or size of objects, merging the appearance of objects in one image with the layout of another, composing objects from many images into one, and more. We also show that self-guidance can be used to edit real images. For results and an interactive demo, see our project page at https://dave.ml/selfguidance/