Introduction of DETECT-World: Advanced Deepfake Detection Model

Resemble AI has launched DETECT-World, a new detection model aimed at identifying deepfakes. This model addresses the challenges posed by the rapid development of AI-generated content, which now includes over 2.2 million variants. DETECT-World combines artifact-based detection with a physics-based understanding of reality, allowing it to evaluate whether content is not only recognizable as fake but also coherent with physical laws. The model is designed to detect both known deepfake signatures and new, unseen manipulations. Its architecture processes video and images in 4-second windows, enhancing its real-time detection capabilities. This dual approach is crucial for combating zero-day attacks, where new deepfake techniques emerge daily. The model is positioned to support various applications, including enterprise voice authentication and media workflows.

Key Points: • DETECT-World combines artifact-based and physics-based detection for deepfakes. • The model processes video and images in 4-second windows for real-time analysis. • It aims to address the rapid growth of AI-generated content, now exceeding 2.2 million variants.