The video model (release video-20261008) now classifies fake videos by attack type, such as text-to-video, image-to-video, face swap, lip sync or inpainting. The attack field is returned by the Detect API and shown on the Detect report.