NVIDIA’s broader push expands AI-enabled verification and workflows, signaling a shift in how media teams approach content authenticity.

The tool is available for commercial and non-commercial use, with guidance to conduct independent testing and caution against uploading sensitive material.

The system is designed for scalable use by newsrooms and broadcasters, integrating into workflows without replacing human editors.

A demo version is available online, but with limitations such as cloud processing, a 100MB file cap, and occasional timeouts under Nvidia’s private access program for media workflows.

The detector is robust to social-media compression and can detect artifacts not easily visible to the human eye, supporting large-scale newsroom use.

Each frame receives a synthetic-likelihood score from 0 to 1, and the tool continues to perform after common processing like compression and re-encoding.

The frame-by-frame analysis feeds into an overall score, guiding editors to escalate potentially synthetic content for review.

NVIDIA unveiled the Synthetic Video Detector as part of its AI for Media platform at SIGGRAPH, a real-time microservice that flags whether video content is real or AI-generated.

The detector analyzes video frame-by-frame, providing a frame-level and overall synthetic-likelihood score to localize and assess authenticity across the footage.

Designed for flexible deployment, it can run on-premises, at the edge, in hybrid setups, or in air-gapped environments to meet security and data-residency needs.

NVIDIA stresses that the detector is a triage tool and not a silver bullet; human verification and traditional fact-checking remain essential.

In testing, the model achieves up to 92% accuracy on uncompressed video, with 87% at 15% compression and 82% at 50% compression; 1080p processing can reach about 22 milliseconds on RTX GPUs and ~30 milliseconds on L40 GPUs.