<p>Reconstruction is an important aspect of gravity and magnetic field data processing. However, existing methods still face issues such as data characteristic recovery distortions, low interpolation and gridding accuracy, and the need for improved denoising capabilities. Diffusion models have been extensively applied and have yielded promising results in image processing tasks. Field reconstruction implemented via the diffusion model is expected to accurately recover characteristics. Therefore, the following work is undertaken. First, a hybrid network model that combines a diffusion model and noise-rating networks is proposed for the simultaneous interpolation, gridding, and denoising of gravity and magnetic field data. Second, through synthetic data tests and comparisons with traditional methods, a validation of the proposed model reveals its high-accuracy reconstruction capabilities. The effects of the number of diffusion steps, number of denoising steps, and degree of data loss on the reconstruction results are discussed. Finally, the model is applied for the reconstruction of real data obtained from the Vinton Dome in the United States. The three-dimensional inversion results obtained for the reconstructed vertical gravity gradient validate the practicality and accuracy of the proposed method. Unlike other methods, such as internal interpolation, blank filling, and edge extending, the developed model has data characteristic recovery capabilities. Compared with the kriging method, the proposed model achieves maximum root mean square error reductions of 22.36% in a blank-filling test and 46.61% in an edge-extending test. Compared with other approaches, the proposed model can more effectively reconstruct gravity and magnetic field data, provide highly accurate data for geophysical inversion, and support geological surveys and mineral explorations.</p>