The human nervous system has the ability to save newly learned movements (i.e. re-learn faster after initial learning) and generalize learning to new conditions. In the context of walking, we rely on savings and generalization of newly learned walking patterns to navigate changing environments and make progressive improvements with gait rehabilitation. Here, we used a split-belt treadmill to study how different perturbation parameters can influence savings and generalization of learning during walking. In Experiment 1, we investigated the effect of split perturbation size on savings of a newly learned walking pattern. We found that larger perturbations led to better savings than smaller perturbations. In Experiment 2, we studied how different features of the initial split perturbation influenced the generalization of learning. Interestingly, we found that practicing the same thing twice did not lead to fastest learning. Instead, initial exposure to larger perturbation ratios led to faster subsequent learning of smaller perturbation ratios as compared to repeated exposures to small perturbations. Collectively, our findings reveal that initial learning conditions can be leveraged to increase savings and shape flexible motor memories during walking.