What kind of technology is used behind this? DO they use models with a lot of logos to train? and how do they generate vector/graphics files?