<p>Microstructural characteristics and particle interactions are pivotal in pharmaceutical manufacturing processes, fundamentally influencing performance such as flowability and dissolution. Variations in microstructural arrangements, even with identical compositions, can contribute to different process performance and product quality. Traditional methods rely on empirical experiments or physical simulations, which are prone to information loss and time-consuming. In this research, an innovative production framework is introduced, including a generative model S2V-GAN (Slice to Volume GAN) with quantitative evaluation metrics and a learning-based flowability prediction model. Specifically, the S2V-GAN consists of a U-net generator and a 2D Slice discriminator using Wasserstein strategy for generating synthetic microstructures that achieve time reduction and rich digital microstructure data through different X-ray Computed Tomography (XCT) scanning scenarios. The prediction model trained with the particle morphology-flow pairing database predicts the flowability of granular medicines through some morphological parameters. The generative model reconstructed the microstructure of single or binary particles by limited training slices and achieved high-level similarity in the distribution of morphological parameters during evaluation. By using the data set to train and fit the flow prediction model, a higher accuracy was achieved in the test, thus realizing the complete process from structure generation to flowability prediction. Finally, the production framework successfully reconstructs and augments high-fidelity three-dimensional microstructures, enabling efficient structural analysis and process optimization.</p> Graphical Abstract <p></p>