English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Plus

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Zendy Tools

Zendy Open

Diffusion models build a new milestone for image generation yet raisingpublic concerns, for they can be fine-tuned on unauthorized images forcustomization. Protection based on adversarial attacks rises to encounter thisunauthorized diffusion customization, by adding protective watermarks to imagesand poisoning diffusion models. However, current protection, leveraginguntargeted attacks, does not appear to be effective enough. In this paper, wepropose a simple yet effective improvement for the protection againstunauthorized diffusion customization by introducing targeted attacks. We showthat by carefully selecting the target, targeted attacks significantlyoutperform untargeted attacks in poisoning diffusion models and degrading thecustomization image quality. Extensive experiments validate the superiority ofour method on two mainstream customization methods of diffusion models,compared to existing protections. To explain the surprising success of targetedattacks, we delve into the mechanism of attack-based protections and propose ahypothesis based on our observation, which enhances the comprehension ofattack-based protections. To the best of our knowledge, we are the first toboth reveal the vulnerability of diffusion models to targeted attacks andleverage targeted attacks to enhance protection against unauthorized diffusioncustomization. Our code is available on GitHub:\url{https://github.com/psyker-team/mist-v2}.

Targeted Attack Improves Protection against Unauthorized Diffusion Customization