Optimal Transport-guided Test-Time Visual Prompting (OT-VP) leverages prompt learning at test time to align target and source domains without altering pre-trained model parameters, achieving significant improvements in accuracy on benchmark datasets.