A Deep Learning Framework for Image Style Transfer and Visual Design Integration in Digital Media Art

Existing models in digital media art often struggle to balance style and content while lacking adaptability to visual design needs. This study proposes an image style transfer framework for visual design integration using a Fully Convolutional Network (FCN) combined with a Visual Geometry Group Network (VGG-Net). The model extracts style features via VGG-Net and applies design constraints using a functional mask preprocessed through the FCN. Experimental results show a content structure retention score of 0.923±0.021 on the training dataset and a style texture similarity score of 0.915±0.032, outperforming benchmark models that reach only 0.851±0.032 in content retention. The proposed method also achieves the lowest frame difference in dynamic style transfer and superior color (8–9) and texture scores (above 8) across five image groups. These results demonstrate that the framework effectively balances artistic style and design intent, providing a robust, knowledge-driven foundation for digital media art innovation.