Prediction of a difficult airway is critical for safe airway management during anaesthesia and emergency. Mallampati score, thyromental distance, and other clinical examination methods have been used commonly; however, these methods have the disadvantage of a lack of precise predictability and high inter-observer variability. Recent developments on computer vision and machine learning have made facial imaging a promising and non-invasive approach to airway assessment. This article reviews the recent works in facial image-based airway prediction, specifically discussing keypoint detection, convolutional neural network, and multimodal fusion of facial features and clinical information. A comparative study of Machine Learning and Deep Learning based methodology with traditional approaches have been presented to aim the efficiency, relevance, and limitations. Additionally, we address the challenges of putting these methods into clinical practice, including insufficient datasets, biases present in algorithms, and the capacity to apply results to diverse populations. This research emphasizes the potential for AI-driven facial image analysis to enhance the precision, consistency, and real-time application of airway evaluations.