Recent advancements in image translation for enhancing underexposed and overexposed images have highlighted the efficacy of deep learning algorithms. Transforming images into high-contrast representations is challenging due to the varying contrast levels within a single image, necessitating models that can adapt to these variations. Deep learning approaches have demonstrated effectiveness in processing RGB images captured under diverse lighting conditions. However, little work exists targeting images with extremely varied exposure with the same image. The existing models do not perform well on such images due to the very nature of the images.In this study, titled HiPyr, we present a novel integration of a HyperNetwork within the Laplacian Pyramid Translation Network (LPTN) to address image enhancement challenges. The HyperNetwork is an innovative architecture that generates weights for another network, allowing dynamic prediction of the optimal kernel function used to decompose images into a Laplacian pyramid. This capability enables a tailored decomposition process that adapts to each image’s unique characteristics. The LPTN is a robust framework that improves image quality through multiscale decomposition and reconstruction, effectively manipulating features at different scales. Our modifications to the LPTN include adjustments to the number of residual blocks and activation functions across all translation network levels. By combining these enhancements with the HyperNetwork’s dynamic kernel prediction, we develop a more efficient and effective image enhancement solution.