At any time of day, there are over 10 million live video streams around the world. However, building video conferencing applications remains challenging for developers, who must manage video bandwidth usage to reduce costs while providing value-added features like face alignment. Researchers at NVIDIA are making this much easier with the Maxine platform SDK. The ultra-low bandwidth Maxine technology saves an initial image of the user’s face and only streams the location of key facial points around the eyes, nose, and mouth. A Generative Adversarial Network (GAN) uses the initial image and key facial points to create a synthetic video, reanimating the subject’s face. The SDK also allows developers to add AI features like real-time translation, closed captioning, and super-resolution. It leverages NVIDIA GPUs to optimize pipelines for high performance.