In the demo's we saw the description being automatically generated from the photo. How does this function work?