The model does not need to see drawings of a horse to produce a picture of a horse. It needs to see pictures of horses, sure, but those could be photographs, drawings, whatever. As a human, you also would not be able to draw a picture of a horse without ever seeing a horse, so I’m not sure what your point is.

Also, how do you know that you’ve had it explained to you well? Unless you’ve attempted to apply the knowledge, you can only tell if you’ve had it explained convincingly