Prediction of complex crack propagation is necessary to guarantee performance in service of critical parts of transport vehicles despite the presence of a crack. In cases where linear fracture mechanics assumptions are not valid, e.g. for ductile rupture, this prediction may require continuous–discontinuous strategies to insert a strong discontinuity (i.e. a crack) into a continuous model based on a scalar variable related to material degradation. These strategies still face difficult challenges regarding robustness and cost-efficiency, in particular for 3D problems. To overcome these challenges, it is useful to consider the different ingredients that constitute a continuous–discontinuous strategy and try to optimize each of them. With this perspective, this work addresses the simulation of crack initiation and propagation in ductile metals and proposes new contributions to some rarely addressed points: (i) a simple geometrical approach to initiate realistic crack shapes in 3D meshes, (ii) a new 3D insertion criterion to minimize the number of transfer operations, (iii) a pragmatic remeshing procedure to refine only Active Process Zones, and (iv) an equilibrium recovery method meant to improve the convergence rate after the insertion of a discontinuity.The effectiveness of these new ingredients is demonstrated in 2D axisymmetric, 2D plane strain, and 3D cases in the case of test specimens used to characterize ductile failure in metals. This includes axisymmetric specimens exhibiting cup-cone fracture, plane strain specimens with slant fracture, and pre-cracked CT-like specimens with crack blunting and large crack propagation.