Runway previews Solaris, a real-time interface world model

Runway previewed Solaris, which it described as the first model in a new category it calls “interface world models” — a system that generates a working software interface one frame at a time, in real time, rather than serving a pre-built application built from code. The model is built on Runway’s Gen-4.5 video generation model, adapted to respond to user actions such as clicks and drags rather than only to a text or image prompt.

Runway said the architecture separates reasoning from rendering: a language model decides how the interface should change in response to input, while the video model renders the resulting frame. The company framed this as avoiding the information loss that occurs when a design is translated into code, allowing an interface to respond more fluidly to what a user does rather than to what a developer anticipated they might do.

Solaris is a research preview rather than a public product — Runway is inviting requests for early access rather than shipping the model — and it is not yet clear from the announcement how reliably the model behaves over extended interaction, a question that has dogged earlier interactive world-model demonstrations from other labs. It extends Runway’s work beyond video generation into rendering interactive software directly, an approach that, if it works reliably, would compete less with other video models and more with the idea of writing application code at all.