This study presents a digital twin-based seam correction framework for collaborative robotic laser welding. The framework addresses key manufacturing challenges, including dimensional inconsistencies, motion inversion, and lightweight part deformation, which significantly impact welding quality. Digital twin technology enables comprehensive analysis and supports real-time deviation monitoring and compensation. Experimental results demonstrate a reduction in maximum seam deviation from 2.9 mm to <0.2 mm, achieving an average deviation <0.1 mm. The system ensures precise laser beam positioning, leading to waterproof welding. Bézier curves are used in the seam correction, providing a flexible and scalable approach. This research contributes to improving welding accuracy and adaptability in flexible production environments while minimizing resource consumption and material waste.