ShowFlow: From Robust Single Concept to Condition-Free Multi-Concept Generation

2025-06-23

AI Summary

Controllable image generation faces two key challenges: (1) in single-concept synthesis, balancing identity fidelity and text-prompt alignment remains difficult; (2) in multi-concept synthesis, reliance solely on textual prompts often leads to identity collapse and concept omission. To address these, we propose KronA-WED—a Kronecker-structured adapter with weight-embedding decomposition—and a decoupled learning framework enabling high-fidelity identity preservation and strong prompt alignment for single-concept generation. For multi-concept synthesis, we introduce Subject-Adaptive Matching Attention (SAMA) and layout consistency regularization, enabling flexible compositional generation without requiring bounding boxes or semantic masks. Our method integrates Kronecker parameterization, decomposition-based adaptation, and attention-level regularization. Extensive experiments and user studies demonstrate significant improvements in visual quality, identity accuracy, and prompt adherence. The approach shows practical utility in real-world applications such as advertising design and virtual try-on.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Customizing image generation remains a core challenge in controllable image synthesis. For single-concept generation, maintaining both identity preservation and prompt alignment is challenging. In multi-concept scenarios, relying solely on a prompt without additional conditions like layout boxes or semantic masks, often leads to identity loss and concept omission. In this paper, we introduce ShowFlow, a comprehensive framework designed to tackle these challenges. We propose ShowFlow-S for single-concept image generation, and ShowFlow-M for handling multiple concepts. ShowFlow-S introduces a KronA-WED adapter, which integrates a Kronecker adapter with weight and embedding decomposition, and employs a disentangled learning approach with a novel attention regularization objective to enhance single-concept generation. Building on this foundation, ShowFlow-M directly reuses the learned models from ShowFlow-S to support multi-concept generation without extra conditions, incorporating a Subject-Adaptive Matching Attention (SAMA) and a layout consistency strategy as the plug-and-play module. Extensive experiments and user studies validate ShowFlow's effectiveness, highlighting its potential in real-world applications like advertising and virtual dressing.

Problem

• Challenges in maintaining identity and prompt alignment in single-concept image generation
• Identity loss and concept omission in multi-concept generation without conditions
• Need for robust framework for both single and multi-concept image synthesis

Innovation

• KronA-WED adapter enhances single-concept generation
• Subject-Adaptive Matching Attention for multi-concept
• Layout consistency strategy without extra conditions