Note that there is no difference in the model or in the training. The only thing needed to convert ios photos into one that IDs people is access to a database mapping name to image. The IDing part is done after the "AI" part, it's just a dot product.