Abstract: In rock physics modeling for oil and gas reservoirs, limited core samples often necessitate generating three-dimensional (3D) digital core samples with varying porosity and water saturation. Conventional mathematical morphology methods fail to achieve precise and continuous control of these parameters, being restricted to discrete values, and are further constrained by a lower limit on water saturation, making it difficult to meet the demand for batch generation of samples with diverse properties. To address these issues, this study introduces a 3D conditional generative adversarial network (3D CGAN). Training samples consist of three-phase digital cores constructed from μ-CT-scanned sandstone cores using mathematical morphology, with porosity and water saturation embedded as conditional variables in the generator. Through adversarial training, the model learns the mapping between microstructural features and target petrophysical properties, enabling batch generation of 3D digital cores with continuously adjustable petrophysical parameters under given structural constraints. Quantitative microstructural comparisons are conducted from four perspectives: pore size distribution, two-point correlation function, pore network, and connected component analysis, with further validation through finite element simulations of resistivity and permeability. Results demonstrate that the proposed method stably generates 3D digital cores with continuously varying porosity and water saturation. The average relative errors are 0.16% (ranging between 0% and 0.4%) for porosity and 0.78% (ranging between 0.25% and 2.0%) for water saturation, indicating that the parameter control accuracy meets the requirements of the study. Comparative results confirm that the generated 3D digital cores exhibit high microstructural similarity to the original cores, with petrophysical responses in good agreement. Even with a single core sample, this method overcomes the lower limit on water saturation and the discrete-parameter restriction of conventional mathematical morphology, providing more accurate and abundant digital cores for rock physics modeling and reservoir parameter estimation.