RealCustom: Narrowing Real Text Word for Real-Time Open-Domain Text-to-Image Customization

TL;DR

RealCustom introduces a novel framework for real-time text-to-image customization that disentangles subject similarity from text controllability, enabling precise and efficient synthesis of images matching specific subjects and descriptions.

Contribution

It proposes a new 'train-inference' decoupled framework with adaptive scoring and mask guidance strategies to improve real-time customization in open-domain text-to-image synthesis.

Findings

Achieves superior similarity and controllability in generated images.

Demonstrates real-time performance in open-domain customization.

Outperforms existing methods in subject relevance and text control balance.

Abstract

Text-to-image customization, which aims to synthesize text-driven images for the given subjects, has recently revolutionized content creation. Existing works follow the pseudo-word paradigm, i.e., represent the given subjects as pseudo-words and then compose them with the given text. However, the inherent entangled influence scope of pseudo-words with the given text results in a dual-optimum paradox, i.e., the similarity of the given subjects and the controllability of the given text could not be optimal simultaneously. We present RealCustom that, for the first time, disentangles similarity from controllability by precisely limiting subject influence to relevant parts only, achieved by gradually narrowing real text word from its general connotation to the specific subject and using its cross-attention to distinguish relevance. Specifically, RealCustom introduces a novel "train-inference" decoupled framework: (1) during training, RealCustom learns general alignment between visual conditions to original textual conditions by a novel adaptive scoring module to adaptively modulate influence quantity; (2) during inference, a novel adaptive mask guidance strategy is proposed to iteratively update the influence scope and influence quantity of the given subjects to gradually narrow the generation of the real text word. Comprehensive experiments demonstrate the superior real-time customization ability of RealCustom in the open domain, achieving both unprecedented similarity of the given subjects and controllability of the given text for the first time. The project page is https://corleone-huang.github.io/realcustom/.