Abstract: Diffusion-based models have recently achieved remarkable success in style transfer. However, when training data is scarce, existing methods struggle to effectively balance style and content.