This model was trained using Unfrozen Model Textual Inversion utilizing the Training with prior-preservation loss methods. There is still a slight shift towards the style, while not using the arcane token.