Efficient, Robust, and Anti-Collusion Fingerprinting of Image Diffusion Models

This work presents a novel fingerprinting method for text-to-image (T2I) models that incorporates anti-collusion capabilities to protect intellectual property. By embedding user-specific identifiers into a personalized normalization module (PNM), the method ensures reliable fingerprint recovery and degrades the quality of images generated by colluded models, achieving over 99.5% fingerprint extraction accuracy. This advancement is crucial for practitioners as it enhances the security of generative models against unauthorized redistribution and improves the robustness of model outputs in collaborative environments.