Abstract: Text-to-image models such as Stable Diffusion, DALL•E, and Midjourney have gained immense popularity lately. However, they are trained on vast amounts of data that may include private, ...