Runway Debuts Project Continuum for Live Video Desktops

Runway Labs has introduced Project Continuum, an experimental operating system shell designed to render user interface regions as real-time, interactive video. The system is powered by Solaris, Runway's first Interface World Model, which is built on its Gen-4.5 video model and builds on its GWM-1 general world model research. Instead of relying on traditional code like HTML, CSS, or document object models (DOM) to define buttons and layouts, Solaris generates interface pixels directly, frame by frame, in response to user actions like clicking and dragging.

The prototype features four distinct interaction modes. 'Portals' generate real-time visual explanations or simulations alongside web browsers. 'Responsive Video Interfaces' dynamically recompose scenes when resized; for example, narrowing a window transforms a straight staircase into a spiral staircase using spatial awareness rather than simple pixel scaling. 'Interactive Worlds' allow users and software agents to navigate simulated environments expanded from static images. Finally, 'Visual Thinking' translates the step-by-step execution of coding agents into a live video stream, offering a visual alternative to text logs.

To achieve real-time responsiveness, Solaris combines autoregressive generation with few-step distillation to target a 720p output with a per-frame latency of under 500 milliseconds. However, Runway has not released a public API, SDK, or pricing, and the technology remains a research preview. For developers, this shift from deterministic code to probabilistic inference introduces significant challenges. There are currently no established methods for testing, state management, or security boundaries. Furthermore, because the interface consists entirely of generated pixels, it lacks the semantic structure required for screen readers and accessibility tools.