Continuum damage models, such as the shear-modified Gurson–Tvergaard–Needleman (GTN) model, describe the progressive loss of stress-carrying capacity due to the nucleation, growth, and coalescence of micro-voids. However, their implementation within finite element frameworks is hindered by pathological mesh dependency and damage localisation. This study extends the non-local GTN numerical model proposed in [1], which incorporates two distinct length scales corresponding to damage mechanisms un der shear-dominated and high-triaxial loading conditions with a particular focus on the transition from normal to shear fracture. An implicit solver is employed in the ABAQUS/Standard, where the material response is updated in a staggered manner. The plastic response is initially integrated using a backward Euler scheme, assuming a constant porosity. This is followed by an update of the non-local porosity field using an integral approach to improve the computational efficiency. The stability of the scheme in the vicinity of the void coalescence regime is also improved by using a viscoplastic regularisation. Although the model uses standard mechanisms for nucleation and growth of the voids, the critical porosity for coalescence is determined using a micromechanics-based criterion. This approach replaces the phenomenological coalescence description of the GTN model with a physically informed mechanism operating at the sub-micron scale. This significantly improves the failure predictions under shear-dominated conditions, where conventional GTN models tend to underestimate the evolution of the damage. The predictive capability, computational robustness, and limitations of the UMAT implementation are evaluated through comparisons with experimental results for low-alloy A533B steel and the corresponding VUMAT simulations. The proposed methodology offers a more computationally efficient and physically motivated framework for modeling ductile fracture across varying stress triaxialities.