The rise of AI-generated media has made creating fake content easier and more realistic, posing societal risks such as disinformation, fake evidence, and fraud. While detection methods for fake media have advanced, they lack interpretability, making them inaccessible and untrustworthy for non-expert users, and limiting their adoption in critical domains. Moreover, most research focuses on detecting fake images, leaving video and emerging 3D rendering formats unexplored. These formats, capable of creating dynamic, multi-perspective fake content, present new challenges that current methods do not address. This project aims to develop a conversational explainability framework for fake media detection across images, videos, and 3D formats. By leveraging large foundation models, users can interact with the system and receive explanations tailored to their expertise. This bridges the gap between technical sophistication and usability, making detection tools more accessible. Additionally, the project adopts a modular approach to interpretability, addressing pixel-level forensics, confidence estimation, and attribution of generative methods. By offering fine-grained transparency at every step, it enhances trust and usability in fake media detection. These advancements have potential for groundbreaking impact in fake media detection and other societally relevant multimedia forensics and computer vision fields, contributing to a safer digital environment.