This model is trained on a dataset of ~650 thematic images, to provide a consistent generalized style without overfitting.