Pool topic: Real-time, continuous, and interactive AI video generation. Question revision: 1. Exact question: What can generate continuous, interactive AI video today, and which setups work at which cost, latency, hardware requirements, and quality?
Checked 9 November 2026
Orbis has a live API. The published median is not viewer delay.
Provider says: Reactor’s Orbis Stable and Dynamic API streams 832×480 video at 18 FPS in 33-frame chunks, with optional upscaling and a WebRTC audio track. A new prompt affects an upcoming chunk. Reactor reports a 1.76-second median processing time per segment after optimization, but no p95 or measured prompt-to-screen distribution in that report.
Three results that should not be combined
Reactor’s integration post says median segment processing fell 39%, from 2.898 to 1.760 seconds. It also says an input change during generation causes the affected upcoming segment to be regenerated. The segment median does not show how often playback stalls.
A separate Google Cloud TPU port reports 1.07× playback generation with audio and 1664×936 upscaling at 18 FPS. The earlier Orbis paper reports 4K at 24 FPS on an optimized GPU serving engine. Neither figure measures this Reactor API’s browser delay or cost.
Check the bill before leaving a session open
Reactor’s billing docs say the meter starts at ready and continues during idle time and recoverable disconnect. The inspected Orbis overview did not show a dollar rate. Before paying for a long run, check the current rate, cap your session, and record a prompt-to-visible response, missed chunks and the billed seconds. We did not run or pay for a session.