ML Security PapersEIRES:Training-free AI-Generated Image Detection via Edit-Induced Reconstruction Error Shift
Wan Jiang 1, Jing Yan 2, Xiaojing Chen 2, Lin Shen 1,2,3, Chenhao Lin , Yunfeng Diao , Richang Hong 1
Published on arXiv
2510.25141
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
EIRES effectively distinguishes real from diffusion-generated images across diverse generative models and remains robust to common post-processing operations without requiring any training
EIRES
Novel technique introduced
Diffusion models have recently achieved remarkable photorealism, making it increasingly difficult to distinguish real images from generated ones, raising significant privacy and security concerns. In response, we present a key finding: structural edits enhance the reconstruction of real images while degrading that of generated images, creating a distinctive edit-induced reconstruction error shift. This asymmetric shift enhances the separability between real and generated images. Building on this insight, we propose EIRES, a training-free method that leverages structural edits to reveal inherent differences between real and generated images. To explain the discriminative power of this shift, we derive the reconstruction error lower bound under edit perturbations. Since EIRES requires no training, thresholding depends solely on the natural separability of the signal, where a larger margin yields more reliable detection. Extensive experiments show that EIRES is effective across diverse generative models and remains robust on the unbiased subset, even under post-processing operations.
Key Contributions
- Key empirical finding: structural edits asymmetrically improve reconstruction of real images while degrading generated images, creating a distinctive 'edit-induced reconstruction error shift'
- EIRES: a training-free AI-generated image detection method that leverages this asymmetric shift, requiring no learned threshold
- Theoretical derivation of the reconstruction error lower bound under edit perturbations, explaining the discriminative power of the shift
🛡️ Threat Analysis
Proposes a novel method to detect AI-generated (diffusion model) images, directly addressing output integrity and content authenticity — classic ML09 AI-generated content detection with a novel forensic technique backed by theoretical guarantees.