Runway Unveils Solaris for Interactive AI Interfaces
Runway has unveiled Solaris, an experimental AI model designed to generate interactive app and website interfaces in real time. The company announced Solaris on August 31, 2026, describing it as the first model in a new family called “Interface World Models.”
Unlike conventional software, Solaris does not rely on a pre-built interface rendered through HTML, CSS or JavaScript. Instead, the model generates the interface itself, frame by frame, while responding to user actions.
Solaris Generates Interfaces in Real Time
Traditionally, developers translate visual designs into code before users can interact with them. However, Solaris removes that intermediate representation and combines rendering with interaction.
When users click, drag or type, those actions become signals for generating the next frame. Meanwhile, a language model interprets the user’s request and determines how the interface should evolve. The Solaris world model then renders the resulting visual state.
Runway says Solaris builds on its Gen-4.5 video model and earlier GWM-1 world model. Furthermore, the system generates frames autoregressively and uses a distilled generation process to achieve interactive speeds. Runway targets a 720p visual output while maintaining coherence throughout a session.
The company demonstrated several potential applications. For example, users could interact with a virtual clothing store by dragging garments onto an image of themselves. Similarly, Solaris could create interactive tutorials and visual demonstrations that respond directly to user actions.
Early Testing Shows Strong User Preference
Runway also tested Solaris against a coded interface generated by Claude Opus 5. The evaluation involved 250 participants across 30 interaction examples, producing nearly 7,500 pairwise judgments.
According to Runway, participants preferred Solaris in 61% of comparisons for following requested interactions. The coded alternative received 24%, while 13% of comparisons were considered equivalent.
Meanwhile, Solaris performed better on natural behavior within a scene. Participants preferred it in 71% of comparisons, compared with 21% for the coded interface. Another 6% rated the results as equivalent.
However, these results come from Runway’s own evaluation. Therefore, independent testing will be important before the performance claims can be treated as established benchmarks.
Text, Accessibility and Reliability Remain Challenges
Despite the potential, Solaris still faces significant technical limitations. Runway identifies stable and legible text as one of its biggest challenges. This matters because text remains central to most software interfaces.
In addition, long-session consistency remains an active research problem. Generated interfaces must preserve layouts, objects and other visual elements over extended interactions. Runway also highlights trust and factual grounding as areas requiring further work.
Accessibility presents another hurdle. Generated interfaces will need to work with screen readers and accessibility APIs to support mainstream software use. Consequently, visual flexibility alone will not be enough for broad adoption.
For now, Runway is working with partners on a public launch and is accepting early-access requests. Thus, Solaris remains a research-stage technology rather than a broadly available software platform.
