Arbitrary image style transfer is a challenging task in computer vision, aiming to transfer arbitrary styles onto content images while maintaining high visual fidelity and flexibility. The primary challenges in this task include ensuring the preservation of fine content details, handling complex style variations, and achieving consistent high-quality results across diverse input images. In this paper, we propose StyleCraft, a unified fusion framework designed to address these challenges. The core innovation of our approach lies in the multi-level fusion of content and style information, which is achieved through an adaptive encoder-decoder architecture, enhanced by ControlNet and AdaIN-based mechanisms. By combining content and style features in a novel way, StyleCraft is able to generate high-quality, visually coherent outputs in arbitrary style transfer tasks. We demonstrate the effectiveness of our model through extensive qualitative and quantitative experiments, where it outperforms existing methods.