A roadmap is presented to transition seamlessly from an image to a predictive computational model for granular materials. So far, constitutive modelling in granular materials has been based on macroscopic experimental observations. Here, the point of departure is the basic granular scale where kinematics, contact forces and fabric control the macroscopic mechanical behaviour of the material. New computational and analytical tools are presented that allow for more accurate measurement of kinematics and inference of contact forces, directly from imaging tools (e.g. high energy tomography). These grain-scale data are then used to construct powerful multiscale models that can predict the emergent behaviour of granular materials, without resorting to phenomenology, but can rather directly unravel the micro-mechanical origin of macroscopic behaviour. The aim of these tools is to furnish a 'tomography-to-simulation' framework, where experimental techniques, imaging procedures, and computational models are seamlessly integrated. These integrated techniques will help define a new physics-based approach for modelling and characterisation of granular soils in the near future.