Visko Deploys Live World Model Orbis on Google Cloud TPUs for Real-Time Streaming

Visko's Orbis Model on Google Cloud TPUs

Visko, an AI startup, partnered with Google Cloud to deploy its Orbis Live World Model on Cloud TPUs. This collaboration focused on enabling real-time streaming of generative video and synchronized audio. The project moved from initial kickoff to production handover within 30 days.

Addressing Real-Time Inference Challenges

Serving a live video world model presents challenges due to the need for continuous, low-latency rendering. Unlike offline video generation, live models require each new segment of video and audio to be rendered before the previous one finishes playing to avoid interruptions. The joint effort between Visko and Google Cloud engineers aimed to overcome this real-time clock bottleneck.

Achieving End-to-End Generation Speed

The teams successfully deployed Visko's end-to-end live inference pipeline, which integrates video generation, super-resolution upscaling, and synchronized audio, onto a single Cloud TPU slice. This optimization resulted in an end-to-end generation speed of 1.07x playback speed, allowing for continuous streaming without freezing.

Redefining Interactive Media

Visko's Orbis model generates continuous video streams from text or image prompts. Viewers can input new steering prompts in real time, dynamically altering the scene without restarting the stream. An example cited is a cat scene where a new prompt introduces a bowl of food, causing the cat to react seamlessly. Orbis launched its public demo on September 1, 2026.