Laser melting deposition (LMD) has the potential to achieve high-performance manufacturing of complex components, however, its process is affected by multiple factors, and ensuring the stability and consistency of the component formation process can be challenging. Existing control schemes include physical model-based and data-driven approaches. Considering the difficulty in establishing accurate mechanistic models to describe the complex LMD process, data-driven methods are employed for controller design. For the layer-by-layer processing principle of multi-layer LMD process, this study proposes a controller based on the Model-Free Adaptive Iterative Learning Control (MFAILC) algorithm. As an intelligent control strategy, MFAILC has the ability to control complex and time-varying systems with repetitive running characteristics and has been applied in various fields. Multiple experiments were conducted on a home-built LMD system, and the thermal imaging camera integrated with the system was used to collect and extract the temperature of the melt pool. To tune the controller parameters, a neural network model based on the experimental dataset was constructed to simulate the process. The designed MFAILC controller is validated by printing experiments based on typical annular thin-walled specimens. Compared with the experiments under constant parameters, the range of deposition height and width of the annular thin-walled structure printed using the MFAILC controller was reduced by 28.6% and 15.4%, respectively, and the MAE was reduced by 31.5% and 48.8%, respectively, and the RMSE was reduced by 33.4% and 35.0%, respectively. The experimental results and comparative analysis show that the designed MFAILC controller can effectively improve the stability of the process and the quality attributes of manufactured parts.