Aim was to identify if the uploaded selfie image was an actual photo or a photo of a photo. A tagged data set of 300,000+ images was created. Various pre-trained Feature Extractor Transformer Learning Models like ResNet152V2, InceptionV3, DenseNet201, Xception, EfficientNetB5 and MobileNet were fine tuned by unfreezing upto top 100 layers. Models were evaluated on metrics like F1-Score, Recall, Precision and AUC. Model performance in terms of latency and system hosting cost were assessed to finalise the model. The final ended up saving over $45,000 per year.