Edge computing has emerged as a popular choice for real-time applications thanks to its low-latency, high bandwidth, and data privacy. In surveillance and autonomous systems operating under low-light conditions, effective image enhancement is essential to improve reliability to make decisions based on the quality of images. However, traditional image enhancement methods often fail to recover sufficient details compared to those from deep learning approaches. Despite advantages in performance, deep learning methods usually have difficult in deployment on edge devices due to their high computational and memory consumption. To addressing this gap, we propose a lightweight and efficient low-light image enhancement framework tailored for the ESP32-S3-EYE. Our design incorporates efficient neural network architectures, model compression, and post-training quantization techniques to achieve real-time performance while maintaining high visual quality. This work demonstrates a practical pathway for deploying advanced deep learning-based image enhancement directly on ultra-low-power edge devices. The experimental results demonstrate our promising outcomes, compared to those reported in previous studies.