DreamMatcher: Appearance Matching Self-Attention for Semantically-Consistent Text-to-Image Personalization

TL;DR

DreamMatcher is a novel method for text-to-image personalization that uses appearance matching self-attention to improve semantic consistency and diversity in generated images, without disrupting the pre-trained model's structure.

Contribution

It introduces a semantic matching-based plug-in approach for T2I personalization that preserves model structure and enhances appearance accuracy.

Findings

Significant improvements in complex personalization scenarios.

Effective preservation of diverse image structures.

Enhanced semantic consistency in generated images.

Abstract

The objective of text-to-image (T2I) personalization is to customize a diffusion model to a user-provided reference concept, generating diverse images of the concept aligned with the target prompts. Conventional methods representing the reference concepts using unique text embeddings often fail to accurately mimic the appearance of the reference. To address this, one solution may be explicitly conditioning the reference images into the target denoising process, known as key-value replacement. However, prior works are constrained to local editing since they disrupt the structure path of the pre-trained T2I model. To overcome this, we propose a novel plug-in method, called DreamMatcher, which reformulates T2I personalization as semantic matching. Specifically, DreamMatcher replaces the target values with reference values aligned by semantic matching, while leaving the structure path unchanged to preserve the versatile capability of pre-trained T2I models for generating diverse structures. We also introduce a semantic-consistent masking strategy to isolate the personalized concept from irrelevant regions introduced by the target prompts. Compatible with existing T2I models, DreamMatcher shows significant improvements in complex scenarios. Intensive analyses demonstrate the effectiveness of our approach.