NVIDIA expands real-time media AI tools at IBC

NVIDIA used IBC 2026 in Amsterdam to broaden NVIDIA AI for Media, adding GPU-accelerated SDKs, NIM microservices, playbooks and blueprints for broadcast, sports, news and streaming workflows. The Synthetic Video Detector NIM microservice now reports 99.3% accuracy for text-to-video content and 97.7% for image-to-video content, with Dalet, TwelveLabs and Wowza integrating it into verification, compliance and live-video intelligence systems.

New and updated tools target production quality and real-time operations. NVIDIA 3D Body Pose turns single-camera video into 2D and 3D motion data for sports analysis, virtual studios and animation workflows, while Video Frame Generation can raise frame rates by 2x or 4x and is being integrated by Ross Video for AI-assisted slow-motion replay. NVIDIA Video Super Resolution, TrueHDR, LipSync, Active Speaker Detection and Studio Voice expand enhancement, HDR conversion, localization and speech cleanup capabilities.

NVIDIA also integrated Media Exchange Layer with Holoscan for Media to help software-defined live production applications exchange video, audio and data across distributed environments. Sports Intelligence Playbooks offer frameworks for fine-tuning open models on proprietary sports footage, with early testing showing multiple-choice accuracy rising from approximately 53% to 94% and open-ended evaluation from approximately 5.7% to 66%. Content Localization technologies are being brought to Holoscan for Media for live captions, translated audio, dubbing, synchronized video and localized graphics.