Despite the rise of digital transactions, paper money remains vital in regions such as Uganda and other parts of Sub-Saharan Africa. The persistent challenge of counterfeit banknotes undermines economic stability and public trust. Traditional detection methods, often reliant on manual inspection, are inadequate against sophisticated counterfeiting techniques. This paper introduces DeepFakesUG, a deep learning system using convolutional neural network (CNN) to identify counterfeit Ugandan banknotes from smartphone images. This study assessed CNN architectures including ResNet50V2, InceptionV3, Xception, and ResNet152V2, and demonstrates that ResNet152V2 and ResNet50V2 excel in detecting subtle counterfeit patterns. Furthermore, we have integrated the ResNet152V2 model into a smartphone application that performs on-device analysis, verifying the authenticity of banknotes in under five seconds without the need for internet access. This study offers a practical, accessible, and efficient alternative for counterfeit currency detection during everyday financial transactions.