Runway unveils Solaris, an AI model that generates interactive app interfaces

Runway, an AI company, introduced Solaris, a groundbreaking AI model capable of generating interactive app interfaces on August 31. Solaris, part of Runway's new family of "Interface World Models", revolutionizes the way we interact with software by creating the interface as users engage with it. Unlike traditional app development, Solaris generates the interface frame by frame, reacting to users' actions.

For instance, an online clothing store could transform into a virtual showroom, enabling users to drag and drop clothing items onto their avatars to preview how they would look. Similarly, a room-design application could allow users to simply instruct the AI to change a sofa's color or move a table, rather than manually selecting options from menus.

Solaris distinguishes itself from traditional AI chatbots by responding visually rather than textually. It could generate a visual scenario if a user requested an interactive demonstration of combustion. This functionality could enhance AI agents' ability to understand and adapt to changing interfaces.

Underlying Solaris is Runway's Gen-4.5 video model and its earlier GWM-1 world model, which enable it to generate the interface one frame at a time. User actions such as clicks and drags serve as inputs to determine what occurs in the subsequent frame. A separate language model assists Solaris in comprehending user intentions and deciding how the scene should transform.

However, Solaris is still experimental and has limitations. One significant issue is the lack of stable, legible text, as the system is based on video-generation technology while software interfaces heavily rely on text. Additionally, concerns about accuracy arise, as an AI-generated interface may appear authentic but contain errors. To address these concerns, Runway is working on measures to ensure Solaris remains grounded in verified product information and can maintain consistency during long interactions.

Despite these challenges, the overarching concept is that an AI-powered operating system could dynamically generate interfaces tailored to individual users and tasks, potentially rendering traditional apps obsolete. This system would generate a unique interface for each user and task, paving the way for a more interactive and immersive software experience.