Made a python script to process danbooru tags into (style) training captions, removing the redundancy, filtering unnecessary tags and converting author names to class tokens.
Saves plenty of tokens to fit into the limits.