Please add a new feature that lets users generate an image conditioned on both a text prompt and an existing image.