NVIDIA’s new AI can detect deepfake videos in just 22 milliseconds

As generative AI becomes increasingly capable of producing videos that are nearly indistinguishable from real footage, the race is no longer just about creating synthetic media. It’s about detecting it before it spreads.

At SIGGRAPH 2026, NVIDIA unveiled Synthetic Video Detector, a new AI-powered verification tool designed to identify AI-generated videos with remarkable speed and accuracy. Rather than replacing traditional fact-checking or forensic analysis, the company says the technology is intended to give newsrooms, broadcasters and enterprises another layer of confidence before synthetic videos enter the public domain.

The announcement comes at a time when deepfake videos are becoming increasingly realistic, making it harder for both people and automated systems to determine what’s authentic. Whether it’s manipulated political speeches, AI-generated celebrity clips or fabricated news footage, synthetic media has rapidly evolved from an internet curiosity into a genuine challenge for journalism, cybersecurity and public trust.

NVIDIA wants AI to fight AI-generated misinformation

Synthetic Video Detector is being introduced as part of NVIDIA’s NIM microservices, allowing organizations to integrate AI-powered video verification directly into existing workflows rather than building entirely new moderation systems.

The system examines videos frame by frame and assigns a probability score indicating whether the footage has been generated or manipulated using AI. According to NVIDIA, the detector can process a 1080p video in as little as 22 milliseconds on RTX systems, making it fast enough for real-time or near-real-time analysis in production environments.

Performance is another headline feature. NVIDIA claims the detector achieves up to 92% accuracy on uncompressed video, with accuracy falling to 87% on videos compressed by 15% and 82% when compression reaches 50%. Compression remains one of the biggest challenges for deepfake detection because platforms like YouTube, TikTok, and Instagram routinely compress uploaded videos, often removing subtle visual artifacts that detection models rely upon.

The company also says the latest version ranks at the top of the AI GVD Bench, an industry benchmark used to evaluate synthetic media detection systems, suggesting it performs competitively against existing open-source and commercial alternatives. The benchmark chart shown in NVIDIA’s presentation highlights the detector outperforming many established models across multiple AI video generators.

Detecting deepfakes is becoming just as important as generating them

The launch reflects a broader shift taking place across the AI industry. Over the past two years, companies have invested heavily in video generation models capable of producing photorealistic clips from simple text prompts.

While these systems have unlocked new creative possibilities for filmmaking, advertising and education, they have also dramatically lowered the barrier to creating convincing misinformation.

For news organizations, the challenge is particularly acute. A single fabricated video shared online during an election, natural disaster or geopolitical crisis can spread globally before human fact-checkers have time to verify its authenticity. That’s why verification tools are increasingly becoming as valuable as the generative models they’re designed to detect.

NVIDIA acknowledges that its detector isn’t a silver bullet. The company says the system is intended to complement existing editorial verification processes rather than replace them. Human oversight, source verification and contextual reporting will remain essential, particularly as generative AI models continue to improve.

Looking ahead, NVIDIA plans to integrate the Synthetic Video Detector into Wowza’s Intelligence Video Framework, making the technology available across more than 35,000 deployments in 170 countries.

As AI-generated video becomes cheaper, faster, and more convincing, the battle against misinformation is entering a new phase. Building better AI is only half the equation. The other half may be building AI capable of telling us when not to believe what we’re seeing.