To overcome the drawbacks of poor robustness of traditional model predictive control parameters and large output current ripple, a visual three-vector model-free predictive control strategy for three-phase three-level inverters is presented. First, the sliding mode observer is used to estimate the lumped parameters of the system, and the reference voltage vector is obtained using deadbeat control. This process does not require any system parameters and improves the robustness of the control strategy parameters. Second, a three-vector combination consisting of redundant vectors is applied during each sampling period to reduce the output current ripple and achieve capacitive voltage balancing. A new visual analysis method is presented, and the effectiveness of the simplified three-vector optimization method is proved in theory. The experimental results show that the proposed strategy can make the grid-connected current of NPC three-level inverters have a lower total harmonic distortion rate when the model parameters are mismatched. This enhances the parameter robustness of the model predictive control.