The model is trained on publicly available images of life in urban China in 1980s. The first training dataset comprises solely of monochrome images. (versions <1.0 are trained on this dataset)

The second training dataset has colored images. (versions >1.0 are trained on this dataset)