Runway unveils Solaris, a model that generates interfaces
Runway has introduced Solaris, a real-time model that renders software interfaces frame by frame instead of running them from code, and is pitching it both as a design tool and as a training environment for agents.
Runway announced Solaris, the first in a new family of what it calls "Interface World Models." The model does not render a fixed app or website built in code. It generates the interface itself, synthesizing each frame as the user interacts with it.
Runway argues that converting a visual design into code limits what software can do and how it looks. In its framing, conventional apps ship as "a lossy compression of the space of possible interactions." Solaris handles rendering and interaction in a single model, with no intermediate representation.
The company highlights three properties. The interface is entirely visual. It is "alive," meaning it renders continuously, so lighting and objects keep changing between user actions. It is also open-ended, supporting behaviors developers never scripted. Runway's examples include a virtual clothing store where users drag a shirt onto a reference photo of themselves, furniture that can be rearranged by voice, and a salad assembled by dragging ingredients into a bowl.
Runway also positions Solaris as a way to train computer-use agents. It notes that today's LLMs struggle with tasks such as booking hotels because they overfit to the layouts they were trained on. Solaris could expose agents to constantly changing interfaces.
Runway names three engineering hurdles: speed, since interactivity breaks down at around half a second of latency; staying coherent over a long session, including text and layout; and cost, since generating every frame is still expensive. Runway's materials also point to a 250-person user study against Claude Opus 5 and to early access for partners only, but the excerpt reviewed gives no further details on results, pricing or availability.