RAIGen: Rare Attribute Identification in Text-to-Image Generative Models

2026-02-06

AI Summary

This work addresses the tendency of text-to-image diffusion models to overlook rare or undefined social, cultural, and stylistic attributes due to biases in training data. To tackle this limitation, the authors propose an unsupervised framework that leverages Matryoshka sparse autoencoders to analyze internal model representations. By introducing a novel metric combining neuron activation frequency with semantic distinctiveness, the method automatically identifies interpretable neurons associated with rare attributes. This approach enables, for the first time, the unsupervised discovery of unknown rare attributes in diffusion models, moving beyond prior methods constrained to predefined fairness categories. Experiments on Stable Diffusion and SDXL demonstrate the framework’s cross-architecture effectiveness, showing its capability not only to audit model biases but also to selectively enhance the generation of underrepresented attributes.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Text-to-image diffusion models achieve impressive generation quality but inherit and amplify training-data biases, skewing coverage of semantic attributes. Prior work addresses this in two ways. Closed-set approaches mitigate biases in predefined fairness categories (e.g., gender, race), assuming socially salient minority attributes are known a priori. Open-set approaches frame the task as bias identification, highlighting majority attributes that dominate outputs. Both overlook a complementary task: uncovering rare or minority features underrepresented in the data distribution (social, cultural, or stylistic) yet still encoded in model representations. We introduce RAIGen, the first framework, to our knowledge, for un-supervised rare-attribute discovery in diffusion models. RAIGen leverages Matryoshka Sparse Autoencoders and a novel minority metric combining neuron activation frequency with semantic distinctiveness to identify interpretable neurons whose top-activating images reveal underrepresented attributes. Experiments show RAIGen discovers attributes beyond fixed fairness categories in Stable Diffusion, scales to larger models such as SDXL, supports systematic auditing across architectures, and enables targeted amplification of rare attributes during generation.

Problem

• rare attribute
• bias identification
• text-to-image generation
• underrepresented attributes
• diffusion models

Innovation

• Rare Attribute Discovery
• Diffusion Models
• Sparse Autoencoders
• Bias Auditing
• Unsupervised Interpretability