text-to-image diffusion models can create realistic images based on input
texts. Users can describe an object to convey their opinions visually. In this
work, we unveil a previously unrecognized and latent risk of using diffusion
models to generate images; we utilize emotion in the inp