The purpose of the present article is to present a dynamic multi-scale approach for turbulent liquid jet atomization in dense flow (primary atomization), together with the possibility to recover Interface Capturing Method (ICM) / Direct Numerical Simulation (DNS) features for well resolved liquid-gas interface. A full ICM-DNS approach should give the best comparison with experimental data, but it is not industrially affordable for the time being, therefore models are mandatory. A numerical representation based on full ICM-DNS, for the initial destabilization of the complex turbulent liquid jet, going up to the spray formation, for which well established numerical models can be used, is appealing but has not yet been applied. Indeed such an approach requires the ICM-DNS to be applied up to the formation of each individual droplet. Hence, in many situation models have to be applied to the dense, unresolved and turbulent liquid-gas flow. To achieve this goal, the most important unresolved phenomena to address are, the sub-grid turbulent liquid flux and surface density, in which models based on the so-called Euler-Lagrange Spray Atomization (ELSA) concept, were developed and have been successfully applied to an Engine Combustion Network (ECN) database, in both RANS/LES (Reynolds-Averaged Navier-Stoke/Large Eddy Simulation) context. An innovative coupling between ICM and a complete ELSA approach was tested based on Interface Resolved Quality (IRQ) sensors to determine locally and dynamically whether or not the interface can be well captured. The ultimate aim is to conduct numerical simulations of fuel injection in an industrial scale, for which comprehensive database has been set up. The test case has been chosen for two reasons: (i) previous numerical studies showed, on the same test case, that RANS turbulence model requires a strong modification to get appropriate results, hence prompted the use of LES models. And (ii), liquid Reynolds and gas Weber numbers are relatively low, compared with ECN test cases, hence more flow regions are expected to be resolved. Results showed that using a fully resolved interface model in the whole domain, provides results in good agreement with the experiment in the primary atomization region only. Indeed, it effectively captured the surface instabilities and liquid structure detachments. In the far field, however, this model becomes rapidly unadapted downward in the dispersed spray region, and the ICM-ELSA model was able instead to treat low volume fractions of atomized liquid, where velocity fluctuations become important.