Runway Unveils Solaris, an AI Model That Renders Software Interfaces Live Instead of Running Code

Runway has unveiled Solaris, a model it describes as the first in a new category called "Interface World Models." Instead of executing code, Solaris generates a software interface frame by frame, in 720p, responding directly to clicks, drags, or voice commands as the interaction happens.

This marks a departure from how software conventionally functions. Normally, an operating system determines what appears on screen, and an application's interface stays fixed until developers ship an update. Solaris instead renders the interface itself in real time, with a language model deciding how the interface should change and the world model producing each frame.

Built on Gen-4.5

Solaris builds on Runway's Gen-4.5 video model and extends work the company began with GWM-1. According to Runway, the system could enable use cases such as:

• Interactive shopping: a store interface that adapts per user, letting shoppers drag a shirt onto a photo of themselves to preview fit
• Product visualization: verbal commands like "move the table" or "change the color of the couch" that instantly alter a rendered scene, aimed at furniture and interior planning
• Visual tutorials: step-by-step demonstrations rendered in context — for example, a combustion demo with different materials — rather than a text-based chatbot answer

Runway also frames Solaris as a potential training environment for AI agents, which it claims currently struggle with tasks like hotel bookings because they're trained on fixed website layouts and fail when interfaces deviate from those patterns.

Runway's larger claim: apps as fixed units could disappear

Runway's broader argument is that the traditional app — a fixed, pre-built unit of software — could become obsolete. Today, completing a task requires opening a specific app: one for shopping, one for messaging, one for reservations. Runway suggests that if an operating system can generate the appropriate interface on demand, there is less need to sort software into a static catalog of apps. An online store, in this vision, would no longer be a fixed layout every user sees identically, but an environment generated and adapted per user.

Still a research project

Runway acknowledges significant unsolved problems. Stable, readable text generation remains difficult, despite being essential to most interfaces. Reliability over long sessions is untested, and Runway itself notes that a wrong but visually convincing result can be worse than no result at all. The company also flags that generated interfaces currently lack support for assistive technology such as screen readers, a gap that would need to be closed before any real-world deployment.

No context window, pricing, or benchmark scores have been disclosed. Runway has not announced a public launch date and is instead soliciting partners and offering early access through a signup form.

What this means

Solaris is a research preview, not a shipping product, and Runway is explicit about that. The idea of interfaces generated on the fly rather than coded and shipped is a genuine departure from decades of software design, but the practical barriers — reliable text rendering, accessibility compliance, and interaction consistency over extended use — are substantial. Whether users or businesses want interfaces that change unpredictably each time they're rendered is also an open question Runway has not answered. For now, Solaris functions as a proof of concept and a signal of where Runway wants to take its Gen-4.5 video technology, rather than a tool ready for production use.