Search papers, labs, and topics across Lattice.
This study investigates the effectiveness of prior-conditioned Gaussian discriminants for detecting AI-generated images, addressing the challenges posed by shifts in generator, prompt/style, and source-domain. By employing a controlled diagnostic approach, the authors assess the value of classifier head training against modern feature separability, revealing that their method often matches or surpasses existing detectors across a comprehensive dataset of 7.1 million images. Key findings highlight the critical role of training prior and the data efficiency of moment-based heads, advocating for improved reporting standards in AI-generated image detection.
Training prior and feature representation can significantly enhance the efficacy of AI-generated image detectors, often outperforming established methods.
Diffusion-based generators have made synthetic images ubiquitous, but detectors often fail under simultaneous shifts in generator, prompt/style, and source-domain. We study AI-generated image detection as a transfer system described by training prior, frozen encoder feature space, and decision rule, and ask when classifier head training adds value beyond what is already separable in modern features. As a controlled diagnostic, we fit a prior-conditioned Gaussian discriminant ladder: closed-form heads built from first- and second-order feature statistics under nested covariance assumptions. On Percept-Lens, a unified protocol over 39 public datasets (7.1 million images), the best rung is frequently competitive with, and sometimes exceeds, released AI-generated image detector heads when matched on both prior and encoder. We further quantify strong sensitivity to the training prior, data-efficiency of moment-based heads, and representation dependence of Gaussian shift metrics, motivating (prior, encoder, head)-level reporting and stronger analytical baselines for AIGI transfer.
