Underwater images are subject to intricate and diverse degradation, inevitably affecting the effectiveness of underwater visual tasks. Traditional deep learning models struggle to operate efficiently on resource-constrained platforms due to their extensive parameter count and high computational demands. Consequently, it is critical to develop lightweight networks for underwater image enhancement. This paper introduces a Lightweight Network with Adaptive Feature Fusion (LAFF-Net) tailored for underwater image enhancement. This network integrates a multi-scale feature aggregation module and a global channel attention module to capture intricate features and enhance feature representation effectively. The streamlined network design, encompassing only approximately 14k parameters, showcases remarkable computational efficiency. Experimental results demonstrate that LAFF-Net significantly surpasses existing methods on real-world underwater image datasets, achieving competitive visual quality performance.