Today’s paper proposes ControlNet++, a new approach to improve the controllability of text-to-image diffusion models when using image-based conditional controls (for example segmentation masks or depth maps).