Software as a Continuous Video Stream
Runway introduced Solaris this week, describing it as the first "Interface World Model": an AI system that generates interactive software interfaces frame-by-frame in real time, without producing or relying on any underlying code. Where a conventional app translates a visual design into code that then renders the interface, Solaris skips that translation step entirely — the interface a user sees is generated directly, frame by frame, as a continuous, responsive video stream. Runway frames the underlying question as: "what happens when an operating system generates apps and websites as you use them?"
How It Works
Solaris pairs a video generation model — built on Runway's Gen-4.5 — with a separate language model that handles decision-making, while the video model handles rendering. User inputs like clicks and drags are treated as conditioning signals that determine what the system generates next, rather than triggering pre-programmed screen transitions. The system generates frames autoregressively with reduced denoising steps for real-time responsiveness, and Runway says it trained the model on its own outputs to help maintain visual coherence across extended interaction sessions. The result, according to Runway, is an interface with no predefined screens or templates — one capable of "alive" ambient motion, responsive physics, and behavior beyond what a developer explicitly anticipated and coded for.
The Case Runway Is Making
Runway argues this approach solves a specific problem for AI agents that operate software: agents trained to navigate conventional, code-based interfaces tend to struggle when a layout changes or presents something unfamiliar, while an interface generated live can adapt to "ever-changing interfaces and never-before-seen layouts" without needing to be re-programmed for each new case. The company points to potential applications including personalized storefronts and tutorials, and frames the technology as raising a bigger question about whether the app, as a fixed, pre-built unit, remains the right basic building block for software at all.
In evaluations Runway published alongside the announcement, human raters preferred Solaris's output over Claude Opus 5's in 61-71% of comparisons for instruction-following and natural scene behavior, and Runway reported that several multimodal language models — including Claude Fable 5 — showed consistent quality degradation when asked to reconstruct complex interfaces from screenshots, a comparison point Runway uses to argue for its direct-generation approach. Runway also acknowledged real limitations: maintaining stable, legible text within generated interfaces, preserving coherence over long sessions, and integrating generated interfaces with existing software infrastructure are all still open problems. It's worth noting that every performance claim in this announcement — the comparison figures, the qualitative advantages — comes from Runway's own evaluations; no independent third-party testing of Solaris has been published yet.
Runway has not announced a public launch date or pricing. The company says it is working with select partners ahead of a broader release and is accepting early-access requests.
What It Means for You
If you build software interfaces or design tools, Solaris is worth tracking less as a finished product and more as a signal of where a segment of the industry is betting attention will go next: generative, code-free interface rendering as an alternative to traditional front-end development. It's genuinely unproven at this stage — text stability and long-session coherence are exactly the kind of unglamorous details that determine whether a technology like this is production-ready or remains a compelling demo, and Runway's own evaluation numbers should be read as a vendor's self-assessment rather than independently verified benchmarks until third-party testing catches up.
