The paper presents a theoretical investigation of compaction banding based on a plasticity model for high-porosity rocks. The selected model is featured by a nonassociated flow rule and two internal variables simulating the competition between softening and hardening in the brittle-ductile transition. The parameters have been calibrated for two extensively studied rocks that exhibited localized compaction under laboratory conditions. In particular, the constants that control the compaction banding domain are defined by matching the stresses at which localized compaction was found in the experiments. The resulting parameters are used to simulate the stress-strain response for two loading modes (i.e., triaxial compression and radially constrained deformation), thus exploring the role of stress paths and kinematic constraints on the evolution of the compaction banding potential. The analyses suggest that the loading paths able to mobilize the plastic resources of the rock alter the potential for compaction banding through irreversible effects. A notable example is oedometric compression, for which the potential to generate localized compaction tends to vanish during the initial stages of inelastic loading. These predictions suggest that constraints to the mode of deformation can hinder the occurrence of compaction bands in the field. Moreover, they suggest that the interplay between kinematic constraints and evolution of the compaction banding potential can be used for experimental characterization purposes. As a consequence, these results emphasize the importance of complementing stress-controlled experiments with other tests able to explore different stress paths, thus providing additional data for an accurate characterization of rheological properties and strain-localization characteristics.