Unsupervised Image-to-Image Translation (I2I) aims to learn mappings from a source domain to a target domain without using paired images for training. In this paper, we propose a novel architecture to accomplish this task. Our goal is to synthesize a multi-domain translation model that is diverse, realistic, and applicable to different domains. However, the generator architectures in existing methods often fail to effectively capture a wider range of image contextual information and extract richer local and global features. The style encoder exhibits limited capability in accurately modeling complex styles, resulting in blurred texture details in the generated images and a lack of accuracy and diversity in the style transferring process. In our work, we explore the role of Omni-Dimensional dynamic convolutions in I2I Translation. We propose a novel generative adversarial network (GAN) model, called OmniStyleGAN, which incorporates an advanced style encoder specifically designed for style-guided image translation tasks. It allows the network to better understand the structure and texture of input images and generate more realistic and higher-quality output images. Extensive qualitative and quantitative experiments on the AFHQ dataset demonstrate the superior performance of our proposed model in achieving better translation results.