Scaling world models and real-time video inference

World models unlock the next frontier of creativity, where creatives can imagine and iterate their vision in real-time.

Join Reactor and Nebius to explore the possibilities of world models and the engineering behind real-time video inference. Learn how Reactor built its custom inference stack, which runs on Nebius, and how managed Kubernetes, self-healing clusters and co-engineering support can help teams bring demanding models to life.

What you’ll learnWhat you’ll learn

• What real-time, interactive video makes possible for media, advertising and gaming applications
• Reactor’s approach to running the fastest inference engine for world models
• Why custom inference kernels and networking matter for running world models on NVIDIA Blackwell
• How Nebius’s GPU infrastructure, managed Kubernetes and failure recovery help keep interactive sessions fast and dependable under load

Who should attendWho should attend

• Technical product leaders evaluating interactive video applications in media, advertising and gaming;
• Application builders and ML engineers exploring video or world models;
• Platform and infrastructure engineers responsible for inference performance and reliability.