Robustness of AI-Art Detectors under Generator Shift

2026-08-12

AI Summary

This study addresses the limited generalization of existing AI-generated image detectors when confronted with emerging generative models such as Stable Diffusion 3.5 Medium. The authors present the first systematic evaluation of five state-of-the-art detectors under zero-shot cross-architecture settings, introducing a novel test set derived from SD3.5 Medium encompassing ten distinct artistic styles. Their analysis integrates CLIP ViT-L/14 embeddings, deep learning classifiers, Grad-CAM visualizations, and reverse prompt engineering. Results reveal that while detectors perform well in-distribution, their accuracy degrades substantially in cross-model scenarios. Among evaluated methods, CLIP ViT-L/14 exhibits the lowest false positive rate but suffers from high false negatives. The findings underscore a significant generalization gap in current detection approaches and advocate for the development of multi-layered defense frameworks against synthetic media.

Abstract

Text-to-image generative models have advanced rapidly, with modern Diffusion Transformer architectures producing images that are increasingly difficult to distinguish from human-created artwork. This development has raised significant concerns regarding copyright protection, misinformation, fraud, impersonation, and the authenticity of digital content. Most AI-art detectors are trained and evaluated on the same generator family, leaving robustness to newer architectures underexplored. In this chapter, we analyze generator shift based on a Stable Diffusion 3.5 Medium (SD3.5m) artwork dataset spanning ten art styles through reverse prompting of held-out human artwork samples. Five detectors are trained on U-Net-based latent diffusion artwork and evaluated in a zero-shot cross-generator setting on the SD3.5m dataset. Deep learning models perform strongly in-distribution but degrade under generator shift, misclassifying many SD3.5m images as human while human false positives remain low. The CLIP ViT-L/14 model performs best overall, while Grad-CAM analysis reveals weaker and more diffuse activation on false negatives. These findings highlight a generalization gap in current AI-art detectors and motivate the development of detectors as one component of a layered defense that remains reliable across rapidly evolving generative architectures.

Problem

• AI-art detection
• generator shift
• robustness
• generalization gap
• diffusion models

Innovation

• generator shift
• AI-art detection
• zero-shot evaluation
• Diffusion Transformer
• robustness