CINEMAE: Leveraging Frozen Masked Autoencoders for Cross-Generator AI Image Detection

2025-11-09

AI Summary

Current AI-generated image detectors suffer from overfitting to generator-specific artifacts, resulting in poor cross-generator generalization. This paper proposes CINEMAE, the first method to adapt contextual inconsistency modeling—previously used in text detection—to the visual domain. It leverages a frozen Masked Autoencoder (MAE) for conditional reconstruction and employs reconstruction uncertainty—formalized as negative log-likelihood—as a transferable, local semantic anomaly signal. Additionally, we introduce a patch-level anomaly scoring scheme and a learnable global–local feature fusion mechanism. CINEMAE requires no fine-tuning and imposes no assumptions about generator architecture or prior knowledge. On the GenImage benchmark, it achieves a mean detection accuracy of 95.2% across eight unseen generative models, significantly outperforming state-of-the-art methods. Its core innovation lies in formulating MAE reconstruction uncertainty as a universal, generalizable detection cue, enabling robust cross-generator detection.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

While context-based detectors have achieved strong generalization for AI-generated text by measuring distributional inconsistencies, image-based detectors still struggle with overfitting to generator-specific artifacts. We introduce CINEMAE, a novel paradigm for AIGC image detection that adapts the core principles of text detection methods to the visual domain. Our key insight is that Masked AutoEncoder (MAE), trained to reconstruct masked patches conditioned on visible context, naturally encodes semantic consistency expectations. We formalize this reconstruction process probabilistically, computing conditional Negative Log-Likelihood (NLL, p(masked | visible)) to quantify local semantic anomalies. By aggregating these patch-level statistics with global MAE features through learned fusion, CINEMAE achieves strong cross-generator generalization. Trained exclusively on Stable Diffusion v1.4, our method achieves over 95% accuracy on all eight unseen generators in the GenImage benchmark, substantially outperforming state-of-the-art detectors. This demonstrates that context-conditional reconstruction uncertainty provides a robust, transferable signal for AIGC detection.

Problem

• Detecting AI-generated images across different generators effectively
• Overcoming overfitting to specific generator artifacts in detection
• Quantifying semantic inconsistencies using reconstruction uncertainty metrics

Innovation

• Uses frozen Masked Autoencoders for semantic consistency
• Quantifies local anomalies via conditional reconstruction likelihood
• Fuses patch statistics with global features for generalization