Towards Building Model/Prompt-Transferable Attackers against Large Vision-Language Models

Open in new window