LaP-Forensics Advances Deepfake Detection with Multimodal Reasoning
LaP-Forensics, a new framework for deepfake detection, combines multimodal reasoning with advanced AI techniques to improve artifact localization and structured evidence referencing. The system uses a Multimodal Large Language Model (MLLM) alongside dual visual streams to enhance the detection of deepfakes. The framework has been tested on various benchmarks, showing significant improvements in identifying manipulated content. The approach focuses on artifact localization, using a combination of RGB semantic streams and DDIM reconstruction-residual streams to detect inconsistencies in digital media.