Edit Away and My Face Will not Stay: Personal Biometric Defense against Malicious Generative Editing

Can we design adversarial perturbations that cause edited images to lose their biometric information, making the edited image biometrically unrecognizable and thereby causing the edit to fail? v We present a novel perspective for protecting personal images from malicious editing by focusing on making biometric features unrecognizable after edits. v We conduct critical analyses on quantitative evaluation metrics commonly used in image editing, exposing their vulnerabilities and the potential for manipulation to achieve deceptive results. v We introduce FaceLock, which incorporates facial recognition models and feature embedding penalties to effectively protect against diffusion-based image editing. v Extensive experiments demonstrate that FaceLock effectively alters human facial features against various editing prompts, achieving superior defense performance compared to baselines.