The model's generation logic has changed, now, using more images actually requires fewer buzz.Does this reduce costs but lower quality?
