I do always wonder what the basis of such machine learning rendering is. There's various traits that we, in modern times, associate with good-looking, successful people, and those traits often come back in face databases because those are commonly based on celebrities.

I can imagine that celebrities in Roman times would look quite different and therefore would produce completely different renders for the same pictures and statues.