Deep Residual Injection for Full-Spectrum Forensic Signal Perception in Multimodal Large Language Models

2026-06-14

AI Summary

This work addresses the challenge that multimodal large language models struggle to simultaneously capture low-level generation artifacts and high-level semantic information for image authenticity verification, which limits their detection performance. The study reveals that semantic representations are primarily established in the early to intermediate layers of such models, and direct fine-tuning often compromises their integrity. To overcome this, the authors propose Deep-VRM, a method that injects artifact signals via residual pathways into intermediate layers, where they are fused with semantic tokens and subsequently modeled by trainable downstream layers. This mechanism adaptively integrates full-spectrum forensic cues while preserving the model’s pretrained semantic capabilities. Extensive experiments demonstrate that Deep-VRM significantly enhances robustness and generalization in detecting AI-generated images across multiple benchmarks.

Abstract

Multimodal large language models (MLLMs) have been increasingly adopted in forensics for their robust semantic understanding. As AI-generated images become realistic, semantic-level inconsistencies alone are often insufficient for reliable detection. This motivates a critical question: whether MLLMs can achieve full-spectrum forensic signal perception, i.e., capturing low-level generator artifacts without sacrificing pre-trained semantic knowledge. We further perform a layer-wise analysis of forensic signal perception in MLLMs, showing that semantic information is primarily formed in the early-to-middle layers, whereas direct fine-tuning for artifact learning disrupts these semantic representations. Based on this insight, we propose Deep Visual Residual MLLM (Deep-VRM) to preserve early semantic processing while injecting artifact-specific visual signals as a residual path into an intermediate layer, where they are fused with semantic token representations and propagated through subsequent trainable layers. This enables later layers to jointly model semantic reasoning and signal-level forensic cues, and surprisingly, the model learns to adaptively leverage different levels of forensic signals depending on the input, achieving robust and generalizable detection performance. Extensive experiments show that our method achieves state-of-the-art across most benchmarks. The code and data are available at https://github.com/KQL11/Deep-VRM.

Problem

• multimodal large language models
• forensic signal perception
• AI-generated image detection
• generator artifacts
• semantic-visual fusion

Innovation

• Deep Residual Injection
• Full-Spectrum Forensic Perception
• Multimodal Large Language Models
• Artifact-Aware Learning
• Semantic-Preserving Fine-tuning