IdentityGuard: Context-Aware Restriction and Provenance for Personalized Synthesis

TL;DR

IdentityGuard introduces a context-aware method for personalized text-to-image models that restricts harmful content related to specific identities while preserving utility and enabling traceability.

Contribution

It proposes a novel, targeted restriction mechanism and watermarking for personalized models, improving safety without sacrificing utility.

Findings

Prevents misuse of personalized models effectively

Maintains model utility and concept integrity

Enables robust traceability of generated content

Abstract

The nature of personalized text-to-image models poses a unique safety challenge that generic context-blind methods are ill-equipped to handle. Such global filters create a dilemma: to prevent misuse, they are forced to damage the model's broader utility by erasing concepts entirely, causing unacceptable collateral damage.Our work presents a more precisely targeted approach, built on the principle that security should be as context-aware as the threat itself, intrinsically bound to the personalized concept. We present IDENTITYGUARD, which realizes this principle through a conditional restriction that blocks harmful content only when combined with the personalized identity, and a concept-specific watermark for precise traceability. Experiments show our approach prevents misuse while preserving the model's utility and enabling robust traceability. By moving beyond blunt, global filters, our work demonstrates a more effective and responsible path toward AI safety.