Skip to main content
TechnologyJul 21, 2026· 2 min read

Just a Frame: Here's NVIDIA's New AI That Detects Fake Videos in Just 22 Milliseconds

NVIDIA has announced the Synthetic Video Detector, a new AI-based tool designed to help distinguish authentic videos from artificially generated ones. The solution will become part of the NVIDIA NIM Microservices platform and has been developed to provide additional support in verifying audiovisual content, especially in contexts where it is crucial to make rapid decisions.

The company emphasizes that advancements in artificial intelligence for video creation have made it increasingly difficult to recognize synthetic content with the naked eye. Although these technologies have numerous legitimate applications, their growing quality may facilitate the spread of footage presented as real, which could have consequences for public trust in information.

To address this issue, NVIDIA has developed a system that analyzes every frame of the video and assigns a probability rating that indicates whether the content was generated by AI. Newsrooms and editorial teams can thus use these results to determine which footage requires additional checks, isolate those considered suspicious, or subject them to more thorough verifications.

Synthetic Video Detector and Data Shared by NVIDIA

The company clarifies that the Synthetic Video Detector does not replace traditional journalistic verification processes. According to data shared by NVIDIA, the system achieves up to 92% accuracy with uncompressed videos. Performance slightly decreases as the compression level increases, stabilizing at 87% with 15% compression and 82% with 50% compression. One of the main strengths concerns the processing speed. The microservice can analyze a 1080p video in about 22 milliseconds on systems equipped with NVIDIA RTX GPUs, while on professional L40 GPUs, the time increases to about 30 milliseconds.

Furthermore, the latest version of the model has also improved results in internal tests, recording higher values for both accuracy and the AUC indicator, which is used to evaluate the system's ability to correctly classify content. NVIDIA also states that the Synthetic Video Detector already ranks among the top in the AI GVD Bench: one of the main industry benchmarks used to assess the performance and accuracy of synthetic video content detection systems and deepfakes.