He knew what the commercial was supposed to be. He knew whether he was drawing from a self-description or a description by a third-party.
So, naturally, he drew more attractive pictures when a third-party was describing the person.
The subject herself, however, is probably socialized not to brag, and so her verbal description filters out the most positive aspects.
On a side note, I'm curious what the results would be if you took someone on the autism spectrum and had them describe a woman they just met. Many such people have no such tact filters; would you get a more accurate drawing from them?
Similar to how getting a description from someone on the autism spectrum would product vastly different results, I'm curious as to how the results of the experience would differ in different countries and different parts of the world.