Performance improvements on the enemy model