Real paired datasets (e.g., RealSR, COZ) face the constraints of high acquisition cost and harsh capture conditions. While existing degradation models aim to simulate the degradation process, a noticeable gap remains between these simulations and real-world degradation. Inspired by style migration, our framework generates low-resolution/high-resolution (LR-HR) paired data that aligns with real degradation patterns by extracting degradation styles from low-resolution images (degradation style carriers) and transferring them to high-resolution images (content carriers). Meanwhile, the contrastive learning method, based on diverse real-world degradations, constructed a negative sample queue to guide the model in concentrating on learning degradation information. Experimental results demonstrate that our framework effectively augments RealSR and other datasets, significantly enhancing the performance of the real-world super-resolution (RWSR) model in practical scenarios.