It appears that the model is capable to generate them both
Maybe I went a little too far with the amount of this character (300 images)
