Enhancing Cross-Prompt Transferability in Vision-Language Models through Contextual Injection of Target Tokens

Open in new window