In the era of generative AI, if you use social media, chances are you’ve seen or watched some AI-generated videos. And with the authentic quality of generated AI content currently making it difficult to tell what is “real”, there is a real need to not only flag but also detect AI content to ensure transparency and minimize the spread of misinformation.
View gallery – 2 images
At the latest SIGGRAPH event currently taking place in Los Angeles, NVIDIA announced: Synthetic Video Detector NVIDIA NIM MicroserviceBuilt for media organizations, newsrooms, and hobbyists. This is an AI-assisted detection model designed to detect AI-generated video content.
Popular now: Ayaneo officially introduces high-end gaming PC pricing for handhelds
According to NVIDIA, this NIM microservice “analyzes the video frame by frame and generates a classifier score for whether it contains synthetic content.” This allows individuals or teams to review and analyze flagged clips and videos. Based on its own internal testing, NVIDIA says its synthetic video detector NVIDIA NIM microservices model achieved detection accuracy of 92% on uncompressed video.
This number drops to 85% for a 15% compressed video and 82% for a 50% compressed video, allowing AI-generated video content to be accurately detected even in the digital age of codecs and artifacts. The benefit for media and newsrooms is that 1080p video can be processed in just 22ms on an RTX system (no GPU specified) or approximately 30ms on an Ada Lovelace NVIDIA L40 enterprise-grade data center GPU.
It’s also designed to scale and integrate with systems, and NVIDIA says it can also handle synthetic video detection in live streaming workflows. One company that is already implementing it at scale is Wowza. The company’s Wowza Video Intelligence Framework is deployed in more than 35,000 locations in more than 170 countries.
FAQ
TweakBot answers frequently asked questions about this news using TweakTown’s own reporting from this page and related content from our archives. Tap a question to see the answer, or enter your answer below.
Question #1
How accurate does NVIDIA’s synthetic video detector NIM microservice achieve for uncompressed and compressed video?
Question #2
How fast can a composite video detector process 1080p video on consumer RTX and Ada Lovelace L40 GPUs?
Question #3
How does the video compression level (15% and 50%) affect the accuracy of the detector?
Question #4
Can NIM microservices be integrated into live streaming workflows at scale?
Have a question not answered here? Ask your question below. TweakBot answers.
“Size matters because many of the organizations most exposed to synthetic media risks, such as broadcasters, government agencies, financial institutions, and critical infrastructure operators, also face stringent requirements for data storage, security, and operational management,” NVIDIA wrote. “Rather than replacing established verification practices, microservices provide another signal for time-sensitive decision-making, helping teams move quickly while protecting editorial standards and ensuring public trust.”
