Image-based pore-scale models are critical for studying pore-scale multiphase displacement physics and connecting pore-scale processes to Darcy-scale constitutive relationships. However, systematic image-based modeling studies remain challenging due to the fragmentation of image acquisition and processing approaches, pore-scale simulators, and post-processing tools. To address this challenge, we develop a human-in-the-loop, AI-assisted workflow that integrates complex image-based modeling tasks into a state-aware framework for multiphase flow simulations. External or local AI tools (i.e., large language models) are used to translate natural-language instructions into validated model actions and parameter updates, while deterministic codes are implemented to execute image generation/import and processing, quasi-static and dynamic two-phase displacement simulations, postprocessing. We demonstrate the framework through deterministic two-phase displacement simulations, direct comparison with reported primary-drainage measurements, and ensemble analyses across porous-medium realizations. Our simulations show that: (1) Sample-scale constitutive relationships (e.g., capillary-pressure–saturation and fluid–fluid-interfacial-area–saturation curves)  from dynamic simulations deviate from their quasi-static counterparts and do not collapse onto universal curves, revealing flow-regime-dependent nonequilibrium behavior. (2) Comparison with reported primary-drainage measurements further shows that model–experiment agreement is strongly affected by pore-body and pore-throat geometric representations. (3) Quasi-static capillary-pressure and interfacial-area relationships are relatively insensitive to microstructural variability, whereas relative permeability is substantially more variable because of its dependence on connected flow pathways. (4) For dynamic displacements, constitutive relationships remain comparatively consistent within each flow regime but distinctly separated among regimes across overlapping saturation ranges. These results highlight the importance of developing regime-aware Darcy-scale constitutive relationships and demonstrate a reproducible framework for model benchmarking, model–data evaluation, and uncertainty-aware pore-scale analysis.