NVIDIA AR SDK
catalog.ngc.nvidia.com
- Free plan
- Windows
- Linux
- Self-hosted
What NVIDIA AR SDK is
NVIDIA AR SDK is a free developer toolkit for adding AI-based face and body modeling to video applications. It uses NVIDIA GPUs with Tensor Cores for high-throughput, low-latency processing. Face detection and tracking provide bounding boxes, while facial landmark tracking predicts 68 or 126 points and head poses. 3D Body Pose Estimation predicts 77 body keypoints in 2D and 3D camera coordinates. Eye Contact estimates gaze angles and can redirect eyes toward the camera within a permissible range. Other features include facial expression estimation, LipSync animation from audio, and Active Speaker Detection using video with synchronized audio. The Core package contains runtime libraries, headers, APIs, documentation, integration guides, installation scripts, and a README. NVIDIA distributes the Core and optional features through NGC, requiring account login for download, and provides GitHub samples plus Linux container instructions.
Who it suits
The SDK is aimed at developers building video-processing applications with AI-powered augmented-reality features. It fits Linux, Windows, and self-hosted deployments using NVIDIA GPU hardware.
Good to know, and what to watch
- Free developer toolkit
- Face and body tracking features
- 68 or 126 facial landmarks
- 77-point 2D and 3D body pose
- GitHub sample applications
- NGC login required to download Core
- Requires NVIDIA GPUs with Tensor Cores
- Windows on Arm has feature restrictions
- Use follows NVIDIA product-specific terms
The verdict
NVIDIA AR SDK offers a broad set of real-time face, body, gaze, expression, audio, and speaker features. Review GPU, platform, licensing, and Windows on Arm constraints before deployment.
What the maker publishes
- Access
- The catalog says users need to be logged in to an NGC account to download the SDK Core. source
- Additional features
- The SDK also includes facial expression estimation, LipSync animation from audio, and Active Speaker Detection using video and synchronized audio. source
- Body pose
- 3D Body Pose Estimation predicts 77 body keypoints in 2D and 3D camera coordinates from video frames. source
- Core package
- The Core package contains runtime libraries, development headers and APIs, documentation, integration guides, feature installation scripts, and a README. source
- Deployment
- The user guide includes instructions for running the SDK in a Linux container. source
- Distribution
- The SDK Core and optional features are distributed through NVIDIA GPU Cloud (NGC), and the Core package provides a script to fetch features from NGC. source
- Eye Contact
- Eye Contact estimates gaze angles and can redirect a person's eyes toward the camera within a permissible range. source
- Face features
- Face detection and tracking uses bounding boxes, while facial landmark tracking predicts 68 or 126 landmark points and head poses. source
- Features
- Features include face detection and tracking, facial landmark tracking, body pose estimation, eye contact, facial expression estimation, LipSync, and active speaker detection. source
- GPU processing
- It uses NVIDIA GPUs with Tensor Cores for high-throughput, low-latency processing. source
- GPU processing
- The SDK uses NVIDIA GPUs with Tensor Cores for high-throughput, low-latency processing. source
- Integration
- NVIDIA provides sample applications demonstrating the SDK features in a GitHub repository. source
- Intended users
- The catalog describes the SDK as a resource for developers building video processing applications with AI-powered AR features. source
- License
- Use is governed by NVIDIA's Software License Agreement and Product-Specific Terms for NVIDIA AI Products; associated models follow the license specified for each model. source
- NIM containers
- The catalog lists Eye Contact and LipSync as available AR SDK features in ready-to-deploy NIM containers. source
- Platforms
- The catalog lists x86-64 Windows and Linux support; the user guide also describes Windows on Arm support with feature restrictions. source
- Purpose
- The SDK provides AI features for real-time modeling and tracking of human faces and bodies from video. source
- Samples
- NVIDIA links sample applications for all AR SDK features from its GitHub repository. source
- Support
- The catalog directs NVIDIA AI Enterprise customers to the NVIDIA AI Enterprise Support Portal and documentation hub. source
- Windows on Arm limit
- On Windows on Arm, LipSync and Active Speaker Detection are unsupported, and only the Body Pose Estimation or Body Detection path is available. source
Checked against the maker's pages on 7 Oct 2026.