Nvidia expands media AI toolkit at IBC with real-time deepfake detection, 3D body pose tracking

Nvidia used the opening of IBC in Amsterdam today to announce a substantial expansion of NVIDIA AI for Media, its suite of SDKs, NIM microservices, playbooks and blueprints designed to inject artificial intelligence into live broadcast, sports and streaming production pipelines without breaking existing environments. The conference, expected to draw more than 44,000 attendees from 170 countries, serves as a primary stage for the media and entertainment industry, and Nvidia is positioning itself as the compute infrastructure behind the next wave of studio automation.

Synthetic video detection reaches prime time

The headline advance is the maturation of Synthetic Video Detector (SVD), a NIM microservice Nvidia first showed at SIGGRAPH earlier this year. According to company data, detection accuracy for text-to-video content has reached 99.3%, and for image-to-video content 97.7%, a sharp improvement in the difficult cases where the source video originates from a static image. SVD does not decide on its own whether material is authentic; it supplies a probability score and frame-level metadata that editorial teams, digital forensics units and media-integrity operations can fold into their existing review processes.

Partners embed the engine into current workflows

Three major partners announced immediate integration. Dalet is embedding SVD into a secure cloud verification workflow for news organizations, letting editors send clips for analysis and receive scores inside their familiar interface. TwelveLabs launched general availability of Compliance by TwelveLabs, the first compliance application on its intelligence platform, which uses SVD for frame-level authenticity signals within the same scanning process that checks against regional and custom standards. Wowza, whose streaming engine powers more than 35,000 deployments across 170 countries, will distribute SVD through the Wowza Video Intelligence Framework; the solution runs on Nvidia-accelerated infrastructure, supports real-time analysis of live feeds, and can be deployed in cloud, edge, on-premises, hybrid or fully air-gapped configurations — flexibility that is critical for broadcasters with strict regulatory and security requirements.

3D motion tracking without mocap suits

In parallel, Nvidia is introducing NVIDIA 3D Body Pose, a technology that extracts 2D and 3D joint positions and angles from single-camera video, no marker-based capture systems required. For sports organizations this means structured motion data for player tracking, biomechanics, performance analysis, enriched replays, officiating workflows, athlete safety and interactive experiences. On the production side, the same data can feed animation blocking, digital doubles, character retargeting and virtual environments, delivering significant time and labor savings versus traditional mocap solutions.

Vizrt already running it in a live virtual studio

Vizrt reports it is already using Body Pose in live virtual-studio environments, with body-motion tracking driving graphics and augmented elements in real time. A demonstration at an event the size of IBC signals the technology has moved from proof-of-concept to genuine production. For the industry, the implication is twofold: a layer of defense against synthetic content that plugs directly into existing workflows, and the democratization of high-quality motion capture without expensive dedicated hardware. Both vectors are pushing media toward a future where AI not only generates content but also verifies, analyzes and structures it, in real time and at global scale.