The model slightly nudges generated images towards late 90s american rural area feel. Training data was not of best quality so use at lower values.