Unleashing the Potential of Vision-Language Models for Generalizable AI-Generated Image Detection

2026-08-05

AI Summary

Existing methods for detecting AI-generated images exhibit limited generalization in complex real-world scenarios and struggle to effectively leverage the high-level semantic information from vision-language models. This work proposes Semantic Prototype Calibration (SPC), which reveals—for the first time—the superior potential of Perception Encoders over mainstream vision foundation models for this task. By constructing and supervising forensic semantic class prototypes, SPC overcomes the representational limitations of conventional linear probing. Built upon a frozen Perception Encoder, the proposed method achieves new state-of-the-art performance across multiple benchmarks, substantially outperforming the DINOv3 baseline and delivering significant gains in detection accuracy, particularly in challenging in-the-wild settings.

Abstract

Recent work has shown that a simple linear probe on frozen representations from modern vision foundation models (VFMs) can achieve state-of-the-art AIGI detection performance, substantially outperforming specialized detectors in challenging in-the-wild scenarios. This finding has established DINOv3 as the dominant foundation-model baseline for subsequent improvements. However, we find that the vision-language model Perception Encoder (PE) holds greater potential for AIGI detection, because its language-aligned representation preserves high-level provenance semantics. Specifically, PE exhibits stronger local provenance organization than DINOv3 in its frozen feature space. However, semantic-agnostic linear probing fails to exploit this structure, as PE-Linear still underperforms DINOv3-Linear by 4.1% on In-the-Wild. Based on this observation, we propose Semantic Prototype Calibration (SPC), which constructs category prototypes from forensic semantic information and calibrates them with supervised data. We apply SPC to PE and refer to the resulting detector as PE-SPC. Our analysis shows that this simple design achieves stronger generalization. Across cross-generator, post-processing, and in-the-wild benchmarks, PE-SPC surpasses the previous DINOv3 baseline and achieves new state-of-the-art results.

Problem

• AI-generated image detection
• vision-language models
• generalization
• provenance semantics
• foundation models

Innovation

• Vision-Language Models
• AI-Generated Image Detection
• Semantic Prototype Calibration
• Generalization
• Perception Encoder