Learning-based Nonlinear Model Predictive Control (LbNMPC) is a promising framework for applying NMPC using dynamics models obtained directly from riding data. We developed a LbNMPC controller based on a black-box dynamics model and compare it with a previously published NMPC controller based on a physics-based description. The controllers have been tested in a high-fidelity simulation framework, and the LbNMPC outperforms the nominal one in terms of tracking indexes, while requiring twice the solution time.