UWM-Net is a semi-supervised deep learning framework I developed to tackle challenges in underwater image enhancement, such as color distortion and detail loss. By combining a Mixture Density Network (MDN) for synthetic data generation with a U-Net for enhancement, the model minimizes dataset requirements while maintaining high performance. The framework combines a Mixture Density Network (MDN) for synthetic data generation and a U-Net for enhancement tasks. The MDN effectively simulates underwater scenes with minimal paired data. The U-Net then utilizes these synthetic images to refine clarity, color balance, and detail in underwater images. UWM-Net achieves competitive results with just 18 pairs of training images, outperforming traditional methods across multiple metrics.