The demos show either totally internal modifications (bouncing blanket changing shape / statue cheeks changing) or isolated with white background images that ha

The demos show either totally internal modifications (bouncing blanket changing shape / statue cheeks changing) or isolated with white background images that have been clipped out. Based on the description of how they generate the splats, I think you’d auto select the item out of the background, do this with it, then paste it back. The splatting process uses a pretty interesting idea, which is to imagine two cameras, one the current “view” of the image, the other one 180 degrees opposite looking back, but at a “flat” mirror image of the front. This is going to constrain the splats away from having weird rando shapes. You will emphatically not get the ability to rotate something a long a vertical axis here, (e.g. “let me just see a little more of that statue’s other side”). You will instead get a nice method to deform / rearrange.