Recognizing vehicles license plates is one of the essential tasks for smart cities. The task of character recognition for license plates is performed by many deep learning state-of-the-art techniques. The challenge comes when the captured license plates images are from complex weather conditions like haze, fog, snow or rain. This paper addresses the challenge of recognizing vehicle license plates in complex weather conditions, such as haze, fog, snow, or rain, which is essential for smart cities. The study focuses on enhancing license plate visibility using state-of-the-art deep learning approaches, particularly generative adversarial networks (GAN) and CycleGAN models. A detailed analysis of cycle GANs generator design, comparing U-Net architecture and autoencoders, alongside deformable and traditional convolution layers are presented. After a comprehensive comparison of these architectures, the paper focuses on character recognition using Python-based OCR libraries. Three libraries—PaddleOCR, Tesseract OCR, and EasyOCR—are evaluated for performance. The results demonstrate that PaddleOCR achieves the best accuracy for recognizing characters from enhanced license plate images. This work provides valuable insights into improving license plate recognition under adverse weather conditions through advanced deep learning models and OCR tools.