The first version, was trained entirely on screenshots from live streams.
Generate the character image of the virtual anchor Qinggua
