Trained with the same data than the previous version, but only with this Colab. I then lowered its weights to 70% with this extension of AUTOMATIC1111. I also decided to use her name instead of a fancy token since she's already vaguely in SD data.