MindBridge is a novel cross-subject brain decoding framework that enables the reconstruction of natural images from brain signals using a single model. The framework addresses three key challenges: size variability in fMRI signals across subjects, diverse neural responses, and limited data availability for new subjects. By leveraging biological-inspired aggregation functions and a novel cyclic fMRI reconstruction mechanism, MindBridge achieves subject-invariant representation learning, enabling efficient adaptation to new subjects. The framework also introduces a reset-tuning strategy to adapt pretrained models to new subjects, significantly improving decoding accuracy with limited data. Experimental results on the NSD dataset demonstrate that MindBridge performs competitively with subject-specific models and achieves high accuracy even with limited data. Additionally, MindBridge enables the synthesis of new brain signals while preserving semantic meaning, opening new possibilities for fMRI data augmentation. The framework's ability to handle multiple subjects with a single model offers significant advantages in terms of efficiency and generalizability, making it a promising approach for broader applications in neuroscience and brain-computer interfaces.MindBridge is a novel cross-subject brain decoding framework that enables the reconstruction of natural images from brain signals using a single model. The framework addresses three key challenges: size variability in fMRI signals across subjects, diverse neural responses, and limited data availability for new subjects. By leveraging biological-inspired aggregation functions and a novel cyclic fMRI reconstruction mechanism, MindBridge achieves subject-invariant representation learning, enabling efficient adaptation to new subjects. The framework also introduces a reset-tuning strategy to adapt pretrained models to new subjects, significantly improving decoding accuracy with limited data. Experimental results on the NSD dataset demonstrate that MindBridge performs competitively with subject-specific models and achieves high accuracy even with limited data. Additionally, MindBridge enables the synthesis of new brain signals while preserving semantic meaning, opening new possibilities for fMRI data augmentation. The framework's ability to handle multiple subjects with a single model offers significant advantages in terms of efficiency and generalizability, making it a promising approach for broader applications in neuroscience and brain-computer interfaces.