No Way To Steal My Face: Proactive Defense Against Identity-Preserving Personalized Generation
IDGuardian uses adversarial perturbations to stop diffusion models from preserving a person's face in generated images.
The paper introduces IDGuardian, a model-agnostic defense against identity-preserving personalized image generation. It adds adversarial perturbations that degrade identity features from external encoders and divert the generative trajectory away from the reference person. The authors say it covers both training-based and training-free personalization and remains effective under post-processing, adaptive attacks, and cross-dataset tests. Code is available at github.com/OpenAscent-L/IDGuardian.
- Defends training-based and training-free personalization pipelines.
- Perturbs both identity extraction and identity injection.
- Reported robust to post-processing and adaptive attacks.
- Code is published at OpenAscent-L/IDGuardian.
- Organizations
- OpenAscent
- AI models
- IDGuardian
From the original article
1 min read at arxiv.orgRecent advances in diffusion models have enabled high-fidelity, identity-preserving image generation for personalized applications such as digital avatars and virtual try-on systems. However, their reliance on sensitive facial reference images raises growing privacy concerns.
By Lizhi Xiong · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.