Designed for portraits
The MODNet model is trained for portrait matting, so it works best when one person is clearly visible from the front or side. Product photos, animals, crowds, illustrations, and objects without a person may produce an incomplete mask.