Runway introduced Solaris on Monday, the first model of what it calls Interface World Models, a new class of AI systems designed to eliminate the conventional process of turning designs into code. Solaris generates interfaces frame by frame in real time, learning from users' clicks, drags, and other interactions to decide what to render next. The company sees this as an early step toward a new kind of operating layer in which interfaces are created on the fly.
Solaris is built on Runway's Gen-4.5 video generation model and observes a user's actions as signals to produce the next frame, learning visually what should happen each time someone interacts with it. Runway reduced the number of denoising steps, trained the model to generate frames autoregressively, and had it learn from its own outputs to keep visual quality consistent over long sessions. The system pairs with a language model that handles reasoning—interpreting user requests, deciding what the application should do next, and creating prompts to guide rendering—while Solaris generates that behavior in real time. Examples Runway offers include dragging a shirt from a rack onto an image of yourself, rearranging furniture in a room, and watching a salad bowl change as you drop ingredients from a pile.
In two evaluations comparing coded and generated interfaces, the company tested whether a coded interface can recreate "the same sense of a living, responsive environment" as an Interface World Model. Runway pitted Solaris against the state-of-the-art language model Claude Opus 5, giving both the same starting image and interaction requests. Across nearly 7,500 pairwise judgments and 30 interaction examples, the 250 participants in its user study preferred Solaris for better adherence to the given instructions in 61% of comparisons and for behaving more naturally within the scene in 71% of comparisons. In a separate test, several state-of-the-art multimodal language models, including Claude Fable 5, were asked to recreate website interfaces from a single screenshot; across 30 interfaces, as visuals became more complex, reconstruction quality consistently dropped, which Runway cites as evidence that translating an interface through an intermediate representation loses information.
Runway says the evaluation results reveal a fundamental difference between coded interfaces and Interface World Models: the former treats each user interaction as an isolated update to the environment, whereas the latter generates interactions that stay coherent within the scene. Because Solaris operates directly on the visual interface, the company claims it can preserve the interface's complete visual and semantic state from the first frame onward, bypassing the implementation step where visual designs are converted into code. The report notes that keeping text stable and legible is a top challenge, along with maintaining coherence over long sessions, grounding generation in richer, verified context, and integrating generated interfaces with the rest of the software stack. Runway says it expects interface generation to follow the development of image and video generation, with successive models improving speed, coherence, and controllability, and envisions a new operating layer that can generate useful interfaces to suit nearly any user need—personalized storefronts and tutorials among them, even questioning whether apps will remain the basic unit of interaction. The company is working with partners to launch Solaris publicly and is accepting requests for early access. If generative interfaces gain traction, the shift could redefine how enterprises think about customization and deployment speed, though it also raises questions about where accountability lies when the interface itself is never explicitly programmed.