Uses the same dataset as V4
2nd version - trained with different dataset from the first version
May create NSFW output