RESEARCHERS at UC Riverside have developed a tool called SAGA (Source Attribution of Generative AI videos) to trace the origins of AI-generated videos. This tool addresses the rising threat of deepfakes and manipulated content by identifying the authenticity, the generative model used, and the development team behind such videos. The motivation stems from increasing instances of deception through deepfake technology, which can lead to severe repercussions like identity theft and misinformation.
SAGA not only detects fake videos but also provides reasoning for their inauthenticity by analyzing temporal signatures, unique artifacts that differ across generative models. This research emphasizes collaboration within the industry to combat fake video proliferation and aims to enhance safety by alerting model creators about misuse of their technologies. The ultimate goal is to prevent the generation of deceptive content altogether.