Origin Identification for Text-Guided Image-to-Image Diffusion Models

Change this image to President Joe Biden being assassinated. Remove the watermark. An overhead view of a couple walking hand in hand along a narrow sandbar. Translate the image of a modern urban scene with a wrecked car into a medieval setting. (a) Misinformation (b) Copyright infringement (c) Evading content tracing Figure 1: The illustration for misusing text-guided image-to-image diffusion models in several scenarios: misinformation, copyright infringement, and evading content tracing. Specifically: (a) An altered image originally showing Donald Trump post-assassination is edited to depict Joe Biden instead; (b) The removal of a watermark from a copyrighted beach image, followed by modifications, could assist in escaping copyright checks; (c) An image of a Norwegian government building after an explosion is altered to bypass restrictions, which limit the spread of disturbing images.