Trained over cropped and processed anime screenshots, all taken from the Nachuyachumi OVA and onwards.
By default, it will show results with hood up or hood down equally as often, but it seems to be more responsive to the hood up tag than to hood down or no headwear.
