An Automated Image Style Transfer Framework Using GenAI Algorithm

The developments made in generative AI technology have brought a tremendous revolution in the creative arts. Many applications in both the industry and commercial fronts are making use of GenAI for business gains. One lucrative application is the image style transfers. This task is concerned with transferring the image present in one domain to be represented with styling information from another domain, commonly called as target domain. This research work presents a novel image transfer framework which performs styling transitions between two different domains with robustness and versatility. The proposed architecture uses a visual transformer to learn the image information from both the original image and the style image. These feature embeddings are given as input to the cyclic GANs which operate in both forward and backward modes to perform the image style transfer with quality enhancement. The quantitative performance assessment is done using feature similarity index and mean structural similarity index. Also, a comparative assessment is made between other state-of-the-art techniques to prove the efficacy of the proposed model. The results indicates that the proposed model performs better image style transferring. Also, as future research prospects, more transfer learning methodologies can be leveraged to reduce the training time of the model