How does that watermark feature work if you can just point the camera at an AI-generated image?

One of the proposals I saw was using the lidar to capture a depth map. You could also embed the GPS location.