Runway has unveiled Solaris, a new class of AI system called an Interface World Model that generates software interfaces frame by frame as users interact with them, potentially eliminating the traditional step of translating visual designs into code. Built on Runway's Gen-4.5 video generation model, Solaris observes clicks, drags, and other user actions to predict what the next screen should look like, making the visual interface itself the application.

How Does Solaris Generate Interfaces Differently From Traditional Software?

Conventional software development requires designers to create visual mockups, then developers translate those designs into code that defines how buttons, menus, and interactions behave. Solaris inverts this workflow entirely. Instead of writing code first and rendering it into a user interface, Solaris generates each frame of the interface in response to user input, learning from interactions to determine what should appear next.

The system pairs a language model with the visual generation engine. The language model interprets what the user is trying to do and decides what should happen next, while Solaris renders that behavior visually in real time. Runway says it optimized the model by reducing denoising steps, teaching it to generate frames autoregressively, and training it on its own outputs to maintain visual quality over long sessions.

To illustrate the concept, Runway offers examples like dragging a shirt from a rack onto an image of yourself, moving furniture around a room, or watching a salad bowl change as you add ingredients. In each case, the interface responds naturally to the action without requiring developers to have pre-programmed every possible interaction.

What Do User Tests Reveal About Solaris's Performance?

Runway conducted two evaluations comparing Solaris to traditional coded interfaces. In the first test, state-of-the-art language models including Claude Fable 5 were asked to recreate website interfaces from a single screenshot. As visual complexity increased, the quality of reconstructed interfaces consistently declined, suggesting that translating an interface through an intermediate code representation loses information.

In the second evaluation, Runway compared Solaris directly against Claude Opus 5, a leading language model. Both systems received the same starting image and interaction requests. Across nearly 7,500 pairwise judgments and 30 interaction examples, 250 participants preferred Solaris in 61% of comparisons for better instruction-following and in 71% of comparisons for behaving more naturally within the scene.

"By doing away with the implementation step where visual designs are translated into code, the image users see quite literally is the application," Runway stated in its announcement.

Runway, company statement

Runway argues that this fundamental difference matters for AI agents. Agents trained on conventional coded interfaces often struggle to generalize when layouts change, but Solaris can enable agents to train on continuously evolving interfaces and never-before-seen layouts.

What Are the Key Advantages and Challenges of Generated Interfaces?

The potential benefits of interface generation extend beyond faster development. Runway envisions a new operating layer where interfaces are generated to suit nearly any user need, including personalized storefronts, dynamic tutorials, and adaptive experiences that respond to individual users. The company even questions whether apps will remain the basic unit of interaction in a world where interfaces are continuously generated.

• Visual Coherence: Solaris maintains coherent visual scenes across multiple interactions, whereas traditional coded interfaces treat each user action as an isolated update to the environment.
• Flexibility and Personalization: Generated interfaces can adapt to individual user needs without requiring developers to pre-build every possible variation or layout.
• Reduced Translation Loss: By operating directly on visual interfaces rather than translating designs through code, Solaris preserves the complete visual and semantic state from the first frame onward.

However, Runway acknowledges significant challenges ahead. Keeping text stable and legible remains a top technical hurdle, as does maintaining coherence over long sessions. The company also notes the need to ground generation in richer, verified context and to integrate generated interfaces with the rest of the software stack.

The reliability question looms large. A button still needs to behave consistently five clicks later, and accessibility, security, saved state, and error recovery must work reliably every time. As one observer noted, the next test will be boring real work, where consistency matters more than the magic of real-time generation.

When Will Solaris Be Available to Users?

Runway says it is working with partners to launch Solaris publicly and is accepting requests for early access. The company expects interface generation to follow the development trajectory of image and video generation, with successive models improving speed, coherence, and controllability over time.

Solaris represents a significant departure from how software has been built for decades. Rather than treating code as the foundation and visual design as the output, Runway is treating the visual interface itself as the primary artifact, generated on demand by an AI system that understands both visual design and user intent. If the approach proves reliable at scale, it could reshape how software is conceived, built, and deployed.