The goal of unpaired image-to-image (I2I) translation is to translate the image from one domain (source) to the image in another domain (target) by merging the content from one source with the style from another. The other state-of-the-art I2I methods often face challenges in consistent and realistic mapping across diverse visual domains while overcoming variations in style, and content. To address it, we have incorporated a particular encoder for style feature extraction, considering its ability to extract global and local style features at different levels. The use of the adaptive instance normalization layer in transferring style makes a flexible and precise integration of content and style features. Further, we have utilized the attention modules in the discriminator to determine if the input image is the real image or the generated image. Experimental evaluation on four benchmark datasets—selfie2anime, horse2zebra, summer2winter_yosemite, and photo2vangogh—utilizing FID and LPIPS scores reveals that the proposed approach works better than other state-of-the-art style transfer methods.