So the low noise model just cleans up the high noise output right? So it doesn't really make sense to apply any LoRAs to the low noise output that aren't the usual speed up ones?