How does the segmentation in YOLO v8 model work?

Question

I know how YOLO models work for object detection.
I wonder what gave the YOLOv8 the ability to apply segmentation at the pixel level.

Is there a clever trick? How does it compare to other models like UNET in its approach?