Traditional digital interfaces rely entirely on operating systems, DOM trees, and precompiled code bases to execute user interactions. Every layout, click handler, and state machine is hardcoded by engineers and remains static until a formal deployment. Runway is proposing an architecture that sidesteps this stack entirely with Solaris, introducing what it categorizes as Interface World Models.
Solaris bypasses standard HTML/JS compilation altogether. Instead, it renders interactive user interfaces frame-by-frame at 720p in real time, reacting dynamically to clicks, drags, and voice inputs. The pipeline pairs a language model—which maps user intent to interface state transitions—with Runway's Gen-4.5 video model to generate the resulting visual feed on the fly, building on the foundation of its earlier GWM-1 system.
Rethinking the Application Layer
By generating interfaces on demand, Runway envisions dynamic generative environments replacing hardcoded application catalogs. In personalized e-commerce, storefronts could morph around individual shoppers, letting a user drag apparel directly onto their own photo for real-time virtual try-ons. For interior design, users can simply state "move the table" or "change the couch color" to produce an immediate visual update.
Beyond retail, the framework targets interactive education by visually demonstrating complex scientific processes—such as material combustion—rather than relying on conversational text explanations. Crucially, this setup also serves as a simulation environment for autonomous AI agents. Modern web agents routinely break when confronted with minor DOM changes; training them within generative world models could force them to generalize across unfamiliar visual layouts.
Structural Hurdles and Research Status
Despite the paradigm shift, the physics of generative video impose steep technical barriers. Generating stable, legible text in a video stream remains a persistent failure mode for diffusion models, making text-heavy enterprise workflows an immediate non-starter. Generating interfaces pixel-by-pixel also demands massive compute per frame—a wildly uneconomic trade-off compared to running lightweight client-side JavaScript.
Furthermore, pure video feeds completely discard semantic accessibility trees, rendering the output entirely incompatible with screen readers. System reliability over extended sessions remains unproven, with visual hallucinations introducing silent failure modes. Runway maintains Solaris strictly as an experimental research prototype rather than an enterprise-ready product. For technical leads, it offers an intriguing preview of UI virtualization, but code-based frontends face no immediate threat.