<p>Residual stresses induced during welding can significantly compromise the structural integrity and performance of welded components. Accurate modeling of the welding heat source is crucial for reliable prediction of temperature distributions and residual stresses. This study introduces an efficient computational method for optimizing welding heat source parameters by integrating computer vision techniques with an improved genetic algorithm. A specialized software, WHSO.2025a, was developed within the Unity platform to facilitate real-time visualization of the molten pool and support unified optimization for multi-layer and multi-pass welding processes. The method incorporates a simplified geometric approach to reduce computational time without compromising accuracy. Additionally, three weld activation methods—birth–death element, field variable, and event series—were evaluated, with the field variable method demonstrating superior computational efficiency. The proposed approach was validated through a case study involving a Q355B single-sided V-groove butt-welded plate with 4 layers and 4 passes. Simulation results closely matched experimental data in terms of molten pool geometry and residual stress distributions, confirming the method's effectiveness and potential for practical engineering applications.</p>