<p>Super resolution has advanced markedly in recent years with the adoption of deep learning techniques. Under ideal conditions, where the degradation process of a low resolution image closely follows a known bicubic downsampling model without unknown factors such as sensor noise or non-ideal point spread functions, supervised deep learning-based SR methods achieve outstanding performance, as they are typically trained on data synthesized using assumed degradation models. In practice, however, real-world low resolution images rarely satisfy these ideal assumptions. To address this limitation, this study proposes a hybrid SR framework that jointly exploits external knowledge learned by a pretrained SR network and internal information inherent to the input image itself. By combining these complementary sources, the resulting hybrid network can adapt more effectively to the characteristics and potential unknown degradations of the given image. Experimental results demonstrate that the proposed approach outperforms state-of-the-art methods, particularly in scenarios where the degradation model is ambiguous.</p>