Physics-Informed Neural Networks (PINN) are powerful tools for solving complex problems combining the predictive power of deep learning with the insights of classical physics. The physical laws are incorporated into the neural network modifying the training process and giving an alternative procedure to solve ordinary differential equations. This methodology has been applied to construct Forming Limit Curves (FLC) in the present work. FLCs are commonly used in automotive or aerospace industry to guarantee the integrity of the component during metal forming processes. The FLC depends on the particular material’s plasticity behaviour, i.e. yield function and work hardening, and on the stress or strain state. A particular strain state can be attained by an infinite number of possible loading histories. However, FLCs are strongly dependent on such histories, which introduces almost an infinite number of possible solutions. PINN provide an extremely efficient approach to consider and predict this non-proportional loading dependence. Typically, FLCs are determined experimentally, numerically, or analytically. An example of the latter is the Marciniak-Kuczynski model (M-K), that has been implemented in a two-step PINN. The first step captures the incremental plasticity, while the second accounts for the strain states generated by arbitrary loading paths. The final output is an optimum neural network that can predict in real time the FLC diagram for any loading path for a given material.