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

The race is no longer just about creating synthetic media, as generative AI can create videos that are nearly indistinguishable from real footage. It is important to detect it before it spreads.

At SIGGRAPH 2026, NVIDIA announced Synthetic Video Detector, a new AI-powered verification tool designed to identify AI-generated videos with incredible speed and accuracy. The company says the technology is not intended to replace traditional fact-checking or forensic analysis, but to give newsrooms, broadcasters and businesses more confidence before synthetic videos enter the public domain.

The announcement comes as deepfake videos are becoming increasingly common, making it difficult for both humans and automated systems to determine what is real. From manipulated political speeches to AI-generated celebrity clips to fabricated news footage, synthetic media has quickly evolved from an Internet curiosity to a real challenge to journalism, cybersecurity, and public trust.

NVIDIA wants AI to fight AI-generated misinformation

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

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

Performance is also one of the key features. According to NVIDIA, the detector achieves up to 92% accuracy on uncompressed video, but accuracy drops to 87% on 15% compressed video, and 82% when compression reaches 50%. Compression remains one of the biggest challenges in deepfake detection, as platforms like YouTube, TikTok, and Instagram regularly compress uploaded videos, often removing subtle visual artifacts that detection models rely on.

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, and shows competitive performance against existing open source and commercial alternatives. Benchmark charts shown in NVIDIA’s presentation highlight that this detector outperforms many established models across multiple AI video generators.

Detecting deepfakes is becoming as important as producing them

This announcement reflects broader changes occurring across the AI ​​industry. Over the past two years, companies have invested heavily in video generation models that can generate photorealistic clips from simple text prompts.

These systems have opened up new creative possibilities for film production, advertising, and education, but they have also dramatically lowered the barriers to producing persuasive misinformation.

For news organizations, this challenge is particularly acute. A single fabricated video shared online during an election, natural disaster, or geopolitical crisis can spread around the world before human fact-checkers can verify its authenticity. As such, validation tools are becoming increasingly valuable as the generative models they are designed to detect.

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

Going forward, NVIDIA will integrate the synthetic video detector into Wowza’s intelligence video framework, making the technology available to more than 35,000 deployments in 170 countries.

The fight against misinformation is entering a new phase as AI-generated videos become cheaper, faster, and more convincing. Building better AI is only half the equation. The other half may be building AI that tells us when not to believe what we see.