Runway introduced Solaris on August 31, 2026, an Interface World Model that generates real-time interactive software interfaces entirely from visuals, eliminating the need for traditional code or intermediate representations, according to Runway. This AI system renders interfaces frame by frame, dynamically responding to user actions.
Solaris represents a significant shift in how applications and websites are built. It bridges the gap between visual design and functional software. The system aims to transform digital experiences into fluid, living environments rather than static pages.
This new approach could change interaction patterns. It allows agents to train against constantly changing interfaces. This improves adaptability compared to models trained on fixed layouts.
How Does Solaris Reshape Software Development?
Solaris fundamentally alters software development by making interfaces entirely visual, continuously evolving, and open-ended. It moves beyond predefined code to create real-time generative experiences. This enables new forms of dynamic user interaction.The system removes the need for a separate implementation step beneath what users see. Imagine a virtual clothing store where the showroom itself functions as the application. Users can drag a shirt onto themselves or rearrange displays as if in a physical store.
Because the application continuously renders, it evolves without waiting for the next user action. Objects respond naturally to manipulation. Users can issue commands like, "Move the table so I can see how it looks," and the scene reacts in real-time.
Solaris supports entirely different behaviors within the same scene. It reacts to user interactions dynamically. This flexibility detaches the interface from predefined workflows.
How Does Runway's Interface World Model Operate?
Solaris functions by adapting Runway's Gen-4.5 video generation model to understand user interaction and respond in real-time. It learns from observed clicks and drags to generate sequential, coherent frames. This system builds on their general world model, GWM-1.
The model treats user input, like clicks or drags, as conditioning for the next frame. It observes interactions as it generates. This helps it infer what should happen next without explicit programming. Frames are generated autoregressively, meaning each frame depends only on the one before it.
Runway converted Solaris into a real-time engine through a three-stage process. They first enabled autoregressive frame generation. Then, they distilled the many-step denoising process into just a few steps. Finally, they trained the fast model on its own outputs to maintain visual quality.
A language model determines how the interface evolves. It interprets user requests and guides Solaris. This separates reasoning from rendering, enhancing responsiveness.
Solaris Versus Coded Interfaces
Solaris outperforms traditional coded interfaces in natural behavior and instruction following. It eliminates the intermediate representation. This results in more coherent and physically grounded interactions.A user study compared Solaris with a state-of-the-art language model. Out of 250 participants, Solaris was preferred in 61% of comparisons for following requested instructions. This was against 24% for coded results. For natural behavior, Solaris was preferred in 71% of comparisons versus 21% for coded websites.
Coded interfaces often produce the requested change. However, they treat interaction as an isolated update. Interface World Models understand object behavior. They generate interactions that feel integrated within the scene.
What Limitations Does Solaris Currently Face?
Current challenges for Solaris include stable text generation, ensuring trustworthiness for critical applications, maintaining coherence over long sessions, and integrating with accessibility tools. These are common hurdles for real-time generative models.
Stable, legible text remains a significant hurdle in video generation. Interfaces heavily depend on text. Trust is also crucial for instructional or commercial experiences. Solaris grounds scenes using real product imagery for reliability.
Maintaining visual and semantic coherence over extended interactions is an active research area. Integration with existing software stacks and assistive technologies, like screen readers, also presents challenges. These areas are expected to improve as underlying models advance.
