You selected 1 image as primary, and then you selected the regions that you would merge. It would retain the background of the primary image, doing what it could to adjust the background according to the secondary's selected area.
If you morphed 2 faces together, it did a pretty decent job. If you put a monkey's head on a zebra's body, well, it wouldn't be so perfect due to the difference in head shape, but for a caricature it was adequate.
On similar images (such as a face), it did a good job of identifying facial features (eye, nose, mouth, & ear positions, for example). It had control points overlayed on each image that you could move to adjust it. So, it showed the 2 images side-by-side, with dots overlayed on what it thought were similar points in each image (such as left/right edges of the mouth & eyes, and top/bottom of the nose & ears, and allowed you to adjust them, before it rendered the merged/morphed image.