NVIDIA Reveals Technology for Detecting Fake Videos at Ultra-Fast Speed

Global company NVIDIA has announced the launch of a new advanced tool powered by artificial intelligence that has the ability to detect and spot fake videos at an ultra-fast speed of just 22 milliseconds, in a step aimed at curbing the rapid spread of visually AI-generated fake content.

The company explained that this innovative technology does not seek to replace traditional journalistic verification processes or digital forensic analysis, but rather has been designed to serve as additional support for media institutions and companies, enabling them to instantly verify the authenticity and validity of videos before approving and publishing them to the public.

The new tool is available as part of the NVIDIA NIM Microservices package, which gives institutions the advantage of integrating it seamlessly and with full flexibility into existing infrastructure systems and workflows, without incurring the costs of building new platforms dedicated to content review.

The tool's working mechanism relies on a frame-by-frame video analysis strategy, and then assigns each clip a numerical probability score indicating the likelihood that it was generated or modified using AI technologies. The tool's efficiency reaches analyzing videos at 1080p resolution in just 22 milliseconds when run using RTX graphics cards, making it the optimal solution for real-time, direct verification within live media production environments.

The company confirmed that its new release succeeded in topping the results of 'AI GVD Bench', the global benchmark specialized in measuring and evaluating the performance of fake video detection systems. The technology outperformed a wide range of competing software solutions, both open-source and commercial, in rigorous tests that included visual materials generated by various and diverse AI models.