Taxonomy of Adversarial Attacks on Text-to-Image Generative Models

We introduce a first of its kind taxonomy of attacks on recent text-to-image generative models, providing a comprehensive understanding of the vulnerabilities of these models to adversarial attacks. We demonstrate the success of these attacks on a popular model and highlight how existing guardrails are insufficient. By identifying potential vulnerabilities and promoting transparency and accountability in the development and deployment of image generation models, this work encourages stakeholders to take proactive steps to mitigate risks and ensure safe and responsible use of these models. This work also highlights the importance of effective defense strategies and can help identify the most robust and reliable models.