Comparison Reveals Commonality: Customized Image Generation through Contrastive Inversion

TL;DR

This paper introduces Contrastive Inversion, a new method for extracting common concepts from small image sets without extra guidance, improving customized image generation quality and fidelity.

Contribution

It proposes a contrastive learning approach to identify and disentangle true semantics of target concepts without relying on auxiliary guidance.

Findings

Outperforms existing methods in concept representation and editing.

Achieves high-level balanced performance in generation quality.

Effectively disentangles true semantics without additional guidance.

Abstract

The recent demand for customized image generation raises a need for techniques that effectively extract the common concept from small sets of images. Existing methods typically rely on additional guidance, such as text prompts or spatial masks, to capture the common target concept. Unfortunately, relying on manually provided guidance can lead to incomplete separation of auxiliary features, which degrades generation quality.In this paper, we propose Contrastive Inversion, a novel approach that identifies the common concept by comparing the input images without relying on additional information. We train the target token along with the image-wise auxiliary text tokens via contrastive learning, which extracts the well-disentangled true semantics of the target. Then we apply disentangled cross-attention fine-tuning to improve concept fidelity without overfitting. Experimental results and analysis demonstrate that our method achieves a balanced, high-level performance in both concept representation and editing, outperforming existing techniques.