This paper presents details of our solution to the CMRxRecon2024 Challenge, where we developed the model, PromptMR \(+\) , for accelerated cardiac magnetic resonance imaging (CMR) reconstruction by utilizing recent advances in deep unrolled neural networks. Our model, consisting of as many as 32 cascades, was trained on approximately 61K slices, covering \(83\%\) of the training dataset provided by the Challenge, which includes raw k-space CMR data with diverse contrasts, anatomical views, and sampling patterns, with acceleration factors ranging from 4 to 24. Submitted under the team CBIM, our approach achieved two first places in both tasks of the multi-contrast and random sampling CMR reconstruction, excelling across all evaluation metrics, including SSIM, PSNR, NMSE, and radiology score. The code is available at https://github.com/hellopipu/PromptMR-plus .