Improving Facies and Property Prediction in Complex Reservoirs Using Enhanced Pix2Pix-Based Modeling

Summary This study presents an enhanced generative AI workflow for reservoir modelling using Pix2Geomodel.v2, an upgraded Pix2Pix-based architecture designed to predict geological and petrophysical properties under increased facies complexity and reduced data requirements. The workflow incorporates a seven-facies geological description and a streamlined 54-layer stratigraphic grid, enabling more detailed representation of heterogeneity while reducing computational effort. Property model(e.g, porosity, permeabilit, and facies) are exported from a reference reservoir model and transformed into paired 2D images for conditional GAN training. The model learns direct relationships between facies, porosity, permeability and volume of shale through property-to-property translation. Preliminary results show that Pix2Geomodel.v2 successfully captures dominant geological patterns, reproduces meaningful cross-property correlations, and maintains spatial continuity consistent with the reference model, as demonstrated by qualitative outputs and variogram analysis. These findings suggest that the enhanced framework can support data-efficient, high-fidelity reservoir characterization, offering a promising path toward reducing modelling uncertainty and accelerating geomodeling workflows