Detection of images in low-visibility poses challenges for safety management of autonomous vessels, coastal surveillance, and maritime situational awareness. Our emotional convolution neural network suffers from performance degradation when raw images are captured under adverse weather, fog and in low-light (dark) illumination. We propose a light-weight Vision Transformer (ViT) based model which provides real-time nautical image enhancement and object detection for lower visibility conditions. We propose a paradigm shift from using a convolution neural network (CNN) as the work-horse of ViTs are good models for capturing long range information but are computationally light-weight and amenable to low-Size Area Power (SAP) embedded systems space applications. Having surveyed and re-used some existing light-weight magic tricks we integrate three ingredients: (1) - a contrast-guided image augmentation module on the basis of dark channel prior, or non-uniform illumination correction techniques [essentially re-establishing “daylight” illumination], (2) a light-weight vision transformer backbone based on Rostrum Rep ViT allowing us to utilize a proportionately lower number of parameters along with the beauty of transformers, and (3) a scale-adaptive feature fusion module for distributing information through cross stage connections. We provide inner workings Anglo-Saxon Esque explanations plus expectations in terms of increasing transparency to underpin the advantageous use of our cherries here. Furthermore, recipe ingredient/Ton berry poison testing using ablation studies confirms miraculously that each of our features/components have “magical” merit. We perform experiments in several data sets RUOD, UTDAC etc and also low-visibility maritime situations (including dense fog). Testing shows a mean Average Precision (mAP) of 94.7% on 321, LOL380 speeds on edge systems via the Py Torch framework. Our model has direct applicability in dense fog and heavy rain amongst other tests, while we show our method achieves required performance with a truly marginal added computational increase 12.8M parameters 18.4 GFLOPs compared to a number of other CANN solutions including YOLOv11 based and RT-DETR. Our work closes the crucial gap in maritime automation that now exists, providing a simple, deployable method to enable real-time object detection on low-resource platforms without sacrificing accuracy, directly impacting autonomous surface vessel (ASV) navigation, maritime port security, fishing vessel monitoring, and climate resilient coastal management systems in keeping with the IMO 2030/2050 decarbonisation targets.