Runway Introduces Solaris, an AI Model for Real-Time Generative User Interfaces

Introducing Solaris: A New Approach to Interfaces

Runway announced Solaris, the initial model in its new category of AI systems called Interface World Models. This system is designed to change how software interfaces are created and used by removing the traditional step of converting visual designs into code. Solaris aims to make the visual interface function as the application itself.

Real-Time Interactive Generation

Solaris operates as a real-time interactive model, generating interface frames continuously. It learns from user inputs such as clicks and drags to determine what to render next. Runway envisions this as a foundational step toward a new operating layer where interfaces are generated dynamically as users interact with them.

Bypassing Traditional Development Steps

Conventional visual interfaces require designers and developers to translate designs into code and pre-define interactions. Solaris is designed to bypass this process entirely. The Interface World Model handles both rendering and user interactions, generating every frame and response to user input, making the entire visual frame the interface. This transforms the interface from a sequence of static pages into a continuously generated, interactive experience.

Underlying Technology and Functionality

Solaris is built upon Runway’s Gen-4.5 video generation model and follows the development path of its general world model, GWM-1. It interprets user interactions as signals to generate subsequent frames, learning visual responses without explicit programming for each interaction. Runway optimized the model by reducing denoising steps, enabling autoregressive frame generation, and training it on its own outputs to maintain visual quality over extended sessions. Solaris also integrates with a language model to separate reasoning from rendering, allowing the language model to interpret user requests and guide the rendering process.