From the InstanceDiffusion paper: "We introduce InstanceDiffusion that adds precise instance-level control to text-to-image diffusion models. InstanceDiffusion supports free-form language conditions per instance and allows flexible ways to specify instance locations such as simple single points, scribbles, bounding boxes or intricate instance segmentation masks, and combinations thereof. Compared to the previous SOTA, InstanceDiffusion achieves 2.0 times higher AP50 for box inputs and 1.7 times higher IoU for mask inputs."