Underwater imagery suffers from various degradation effects due to the physical properties of water, including light absorption, scattering, and color distortion. These challenges significantly impact applications in marine research, underwater robotics, and ecological monitoring. While generative models like GANs and VAEs achieve state-of-the-art performance in underwater image enhancement, they suffer from issues like mode collapse and vanishing gradients. Diffusion models have shown superior performance in image generation and restoration tasks, but have been computationally expensive due to the iterative nature of the method, making them unsuitable for real-time underwater vision tasks. This paper presents ClearOcean, a novel Pyramid Diffusion model for efficient underwater image enhancement. This approach combines multi-scale denoising with spatial and channel-wise attention modules, achieving state-of-the-art performance while reducing computational requirements. A smaller model, ClearOcean-S, is also proposed to balance performance and inference speed. Both models were evaluated on the LSUI and UIEB datasets, demonstrating significant improvements in both image quality and processing speed compared to existing methods. Additionally, real-time inference is achieved by the ClearOcean-S model, making it suitable for real-time inference requirements.