we can conclude that the generated images have some similar characteristics with the retrieved training images but are essentially different