Digital image authentication has become increasingly important with the ready availability of online image editing tools and the spread of manipulated content on social media. Conventional forgery detection systems often fail to localize tampered regions or provide interpretable explanations. In this paper, we introduce a strong and explainable model that combines Error Level Analysis (ELA), and a Vision Transformer (ViT) encoder-decoder component. ELA identifies areas of possible forgery by examining compression artifacts. These are embedded and fed into a ViT encoder to capture global dependencies essential for spatial forgery detection. The cross-attentive decoder creates natural language descriptions outlining the reason and whereabouts of tampering. It is an end-to-end approach not just for classifying and localizing regions of forgeries but even creating sentence-level explanations, bringing higher transparency in digital forensics. Experiments on benchmark dataset “CASAI V1” and “CASAI V2”, proves stronger detection capabilities over weak and subtle forgery, for authenticating content-based applications in reality.