Image super-resolution (SR) has garnered significant attention in recent years due to its potential applications in various domains such as medical imaging, surveillance, and satellite imagery. SwinIR, a state-of-the-art deep learning model, has shown remarkable performance in SR tasks by leveraging hierarchical global information modeling. However, to further augment SwinIR’s representational capabilities, we introduce SwinAGI, an enhanced variant incorporating additional mechanisms for refined global information modeling. Specifically, by substituting SwinIR’s traditional 3 \(\,\times \,\) 3 convolution operation with the Self-Calibrated Enhance Block (SCEB), SwinAGI enhances the capture of global information. Moreover, a novel multilayer perceptron (MLP-mixer) is introduced to fine-tune SwinIR’s attention mechanism. Experimental results on prominent test sets demonstrate a significant enhancement, highlighting the effectiveness of SwinAGI in advancing the state-of-the-art in SR performance.