It was primarily trained on photos of Eua, but was trained on some images of Featherine. I recommend lowering the detail of the LoRa to 0.9~0.7.