Download Model for each video run

That’s nice, but how do you do that exactly?