A text prompt for guiding the model during image segmentation. Required for prompt mode and semantic mode, disallowed for other modes.