We perform deep extraction of EXIF, XMP, and IPTC metadata to identify inconsistencies in software signatures, timestamps, and GPS coordinates that may indicate tampering.
For audio and video files, our system analyzes spectrograms to detect "splices" or frequency shifts typical of AI-generated voices (ElevenLabs, RVC) and deepfake sync issues.
Our algorithms search for "checkerboard artifacts" and pixel-level inconsistencies generated by GANs (Generative Adversarial Networks) used in DALL-E, Midjourney, and Sora models.