Abstract: To solve the disability of modeling inter-domain semantic and intra-domain long-range information in current image style transferring algorithms this article proposes a novel image style transferring algorithm SSC-GAN.This method extracts the semantic features by constructing the semantic shortcut connections,thereby enhancing its capability of modeling semantic difference between image domains.Meanwhile,the self-attention mechanism is introduced for the modeling of long-range dependency in the image domain.The SSC-GAN can significantly improve the performance of image style transferring without extra computation.Through the extensive experiments on vangh2photo and selfie2anime datasets,the proposed method achieves excellent visual effects and reduces FID and KID by 1.3 and 1.1 on average respectively,which verifies the effectiveness of SSC-GAN.