We translate grasps between different robot hands using vision and Schrödinger Bridges, enabling generalizable, stable grasp synthesis without paired data or hand-specific retraining.