MultiLayerFakeDetect - Multilayer multimodal analysis for reliable detection of AI-generated and manipulated content across documents, images, audio, and video

MultiLayerFakeDetect develops a multi-layer, multimodal approach for the reliable detection of AI-generated and manipulated content across documents, images, audio, and video. Modern generative AI makes it increasingly easy to create convincing fakes — from manipulated photos and cloned voices to fully synthetic videos. The project combines analyses at several levels (content, signal, and metadata) and across modalities to robustly identify manipulations and synthetic media. A particular focus lies on quality-adaptive thresholds and on detecting inconsistencies such as audio-video desynchronisation. The goal is a practical solution for safeguarding media integrity. The project is funded by the FFG Basisprogramm.