SecureGen: A Robust Framework for Enhancing Security and Resilience in Text-to-Image Generation Models

Text-to-image generation models have made progress in recent years. However, their use in secure environments is limited due to issues like vulnerability to adversarial attacks and difficulties with detection systems. In this paper, we introduce SecureGen, a framework designed to address these challenges while maintaining high-quality image generation. SecureGen includes a multi-layered attention mechanism, dynamic prompt adjustment, multilingual support, and a security post-processing fusion mechanism. These components help SecureGen improve its ability to avoid detection systems, adapt to various tasks, and generate high-quality images. The design of SecureGen ensures both security and robustness in generative models, offering a direction for safe deployment in high-security applications.