TranX-Adapter: Bridging Artifacts and Semantics within MLLMs for Robust AI-generated Image Detection

2026-02-25

AI Summary

The photorealism of AI-generated images poses a significant threat to information authenticity, yet existing detection methods suffer from attention dilution when fusing texture artifacts and semantic features within multimodal large models, limiting their performance. To address this, this work proposes TranX-Adapter, a lightweight adapter that introduces, for the first time, an optimal transport mechanism using a cost matrix constructed via Jensen–Shannon divergence, coupled with task-aware X-Fusion cross-attention to enable efficient bidirectional transfer between artifact and semantic features. This approach effectively mitigates attention dilution and substantially enhances feature fusion quality, achieving consistent performance gains across multiple AI-generated image detection benchmarks, with accuracy improvements of up to 6%.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Rapid advances in AI-generated image (AIGI) technology enable highly realistic synthesis, threatening public information integrity and security. Recent studies have demonstrated that incorporating texture-level artifact features alongside semantic features into multimodal large language models (MLLMs) can enhance their AIGI detection capability. However, our preliminary analyses reveal that artifact features exhibit high intra-feature similarity, leading to an almost uniform attention map after the softmax operation. This phenomenon causes attention dilution, thereby hindering effective fusion between semantic and artifact features. To overcome this limitation, we propose a lightweight fusion adapter, TranX-Adapter, which integrates a Task-aware Optimal-Transport Fusion that leverages the Jensen-Shannon divergence between artifact and semantic prediction probabilities as a cost matrix to transfer artifact information into semantic features, and an X-Fusion that employs cross-attention to transfer semantic information into artifact features. Experiments on standard AIGI detection benchmarks upon several advanced MLLMs, show that our TranX-Adapter brings consistent and significant improvements (up to +6% accuracy).

Problem

• AI-generated image detection
• artifact features
• semantic features
• attention dilution
• multimodal large language models

Innovation

• TranX-Adapter
• Optimal-Transport Fusion
• Cross-Attention
• Artifact-Semantic Fusion
• AI-generated Image Detection