Recent advances in deep learning have greatly improved image deblurring for natural scenes. However, applying these methods to physical experiments, especially those involving rapid, complex dynamics like hydrodynamic instabilities, remains challenging. Unlike conventional deblurring tasks, these scenarios involve motion blur tied to evolving physical processes, complicating image restoration. We propose ClariPhy, a transformer-based approach utilizing the Restormer model, fine-tuned on a novel deblurring dataset derived from Rayleigh-Taylor instability simulations. This dataset features pairs of sharp and artificially spatial and temporal blurred images, reflecting the real-world conditions of physical experiments. Leveraging the self-attention mechanism of transformers, ClariPhy effectively captures spatiotemporal dependencies crucial for deblurring images of dynamic phenomena. Our results show that ClariPhy outperforms the original SOTA Restormer model, providing enhanced clarity and accuracy in time-sensitive physical experiments. Source code and dataset: https://doi.org/10.5281/zenodo.13385382.