Nvidia launches AI tool to detect fake videos at high speed
WASHINGTON, July 21 (QNA) – NVIDIA announced a new AI-powered tool that can detect manipulated or AI-generated videos in as little as 22 milliseconds, with the goal of curbing the rapid spread of synthetic visual content.
The company said the new technology is not a replacement for traditional journalistic verification or digital forensic analysis. Instead, it is designed to serve as an additional verification tool for media organizations and businesses, allowing them to quickly assess the trustworthiness and trustworthiness of videos before publishing them.
The tool is available as part of the NVIDIA NIM microservices suite, allowing organizations to seamlessly integrate it into their existing infrastructure and operational workflows without having to build a dedicated content verification platform from scratch.
The detector analyzes the video frame by frame and assigns each clip a probability score that indicates the likelihood that it was generated or modified using AI. When running on NVIDIA RTX GPUs, it can process 1080p video in just 22ms, making it ideal for real-time verification in live media production environments. NVIDIA also stated that this model remains valid even after common processing steps such as compression, resizing, cropping, and re-encoding.
According to the company, the new detector ranked #1 on AI GVD Bench, a benchmark for evaluating AI-generated video detection systems, and broadly outperformed both open source and commercially available alternatives in tests on videos created by a variety of generative AI models. (QNA)
