Runway Tests AI Interfaces That Draw Themselves in Real Time

Runway has unveiled Solaris, an AI system that renders software interfaces live instead of running them as traditional code. The idea could make shopping, product visualization and tutorials more adaptive, but readable text, reliability and screen reader support remain unresolved.

WTF Index IDIOCRACY
◄ Terminator 1 Idiocracy 2 ►

Generated interfaces could make software more adaptive but raise reliability, readability and accessibility concerns that mildly point toward lower quality and dependence.

Runway Tests AI Interfaces That Draw Themselves in Real Time

Runway is testing a different idea of what software could look like. Its new Solaris system does not treat an interface as a fixed app screen produced by code. Instead, it generates the interface visually, frame by frame, while a person interacts with it.

The company calls Solaris the first model in a new category of "Interface World Models." It remains a research effort, but the concept points toward a future where user interfaces are less like static products and more like generated environments.

How Solaris Changes the Interface

Traditional software depends on code, operating systems and app updates to define what appears on screen. A button, page or menu generally behaves according to rules built into the application. Solaris takes a different route: it renders the interface itself in 720p and responds directly to clicks, drags or voice commands.

That means the system is not simply turning a design into code before a user sees it. The interface is generated live as the interaction unfolds. A language model decides how the interface should change, while the world model renders each frame.

Solaris builds on Runway's Gen-4.5 video model and follows work the company began with GWM-1. In this setup, the boundary between video generation and software interface design becomes much less clear. The screen is no longer only a place where software is displayed; it becomes the generated software environment itself.

Where Runway Thinks It Could Be Useful

Runway's examples focus on situations where a fixed layout can feel limiting. Online shopping is one of the clearest cases. Instead of navigating a standard store page, the store itself could become the interface.

In that version of shopping, a user could drag a shirt onto a picture of themselves to try it on. The store would not have to present the same layout to everyone. It could adapt to each user and to the task happening in the moment.

Product visualization is another area Runway highlights. A person could say "move the table" or "change the color of the couch," and the scene would respond. Runway sees that as relevant to furniture and interior planning, where the value comes from seeing changes in context rather than reading about them.

The company also points to tutorials and learning. Instead of receiving a text answer from a chatbot, a user could see the next step visually inside the generated environment. Runway gives the example of a combustion demo with different materials, where the explanation is shown as part of the scene.

The Bigger Bet Against Fixed Apps

Runway's larger claim is that the "app" as a fixed unit could eventually disappear. Today, users usually choose the right prebuilt app for the job: one for shopping, one for messages, one for reservations. Solaris suggests a different pattern, where an operating system could generate the interface needed for a task without relying on a fixed catalog of apps.

That would be a major shift in how people think about software. Instead of opening a static tool and adapting to its layout, the interface would adapt around the user and the goal. A store, tutorial or planning tool could be generated as a custom environment rather than presented as the same screen for everyone.

Runway also sees Solaris as a possible training ground for AI agents. The company says agents still fail at tasks like hotel bookings because they are trained on fixed layouts and struggle when websites depart from those patterns. Generated interfaces could give agents more varied environments to learn from.

Why Solaris Is Still Early

The idea is ambitious, but Runway is clear that important problems remain. Stable, readable text is still difficult, even though interfaces depend heavily on text. If labels, instructions or options are unreliable, the usefulness of a generated interface drops quickly.

Long sessions and reliability are also open questions. A generated screen that looks convincing but is wrong can be worse than showing nothing at all. For software interfaces, appearance is not enough; the result has to remain dependable across interaction.

Accessibility is another unresolved requirement. A generated interface must work with assistive technology such as screen readers. Without that, the system would leave out users who depend on those tools to navigate software.

For now, Solaris is not being presented as a finished replacement for apps. Runway is looking for partners for a launch and offers early access through a form. The research effort is less a finished product than a test of whether AI software interfaces can move beyond fixed screens and toward live, adaptive environments.