📖 Make the Most of Everything

Cite as: arXiv

Submitted on 2025/03

REF: https://arxiv.org/abs/2503.13945

Comparison across different black-box prompts

Anti-Dreambooth (Anti-DB)

Alternating Surrogate and Perturbation Learning (ASPL) to approximate the real trained models and alternately performs Dreambooth training and attack.

SimAC

Leverages a greedy algorithm to select timesteps with the highest gradient scores to update the adversarial example.

DisDiff

Set cross-attention erasure loss to erase the keyword’s attention in attacking the Dreambooth process.