Runway has unveiled Solaris, the first model in a new category the company calls "Interface World Models" — AI systems that generate user interfaces live, frame by frame, instead of running code. Rather than translating a design into an app, Solaris renders the interface itself in 720p and responds directly to clicks, drags and voice commands as you interact with it.
The announcement, covered in detail by The Decoder, marks a sharp break from how software normally works. Today, an operating system decides what happens on screen, and an app stays the same until a developer ships an update. Solaris proposes the opposite: an interface that exists only as long as it is being generated and adapts continuously to the person using it. For more context on this story, see our ongoing AI industry coverage.
How an Interface World Model Works
Solaris is built on Runway's Gen-4.5 video model and follows the direction the company started with GWM-1, its earlier general-purpose world model. Under the hood, two systems cooperate: a language model decides how the interface should change in response to what the user does, while the world model renders each frame.
The result looks and behaves like an app, but nothing is running in the traditional sense. There is no layout file, no front-end code, no compiled binary — just a generative model producing the next frame based on your last click.
The Use Cases Runway Is Pitching
Runway lists several scenarios where a generated interface could replace a fixed one.
In online shopping, the store itself becomes the interface. Instead of browsing a static product grid, a shopper could drag a shirt onto a picture of themselves to try it on. The shop would no longer be a fixed layout that every visitor sees the same way, but an environment that adapts to each user.
For product visualization, the pitch is conversational manipulation of objects: say "move the table" or "change the color of the couch," and the scene reacts instantly. Runway sees this fitting furniture shopping and interior planning, where seeing a change beats imagining it.
The company also points to tutorials and learning. Instead of a chatbot answering with text, Solaris could show the next step visually in context — the company's example is a combustion demonstration that changes depending on the materials involved.
Runway's Bigger Claim: The App Disappears
Underneath the demos is a provocative thesis. Runway argues that the "app" as a fixed unit of software will disappear. Getting something done today means opening the right prebuilt application — one for shopping, one for messages, one for reservations. If an operating system can generate the right interface on demand, the reasoning goes, there is far less reason to sort software into a fixed catalog.
There is a secondary audience for the technology: AI agents themselves. Runway notes that agents still fail at tasks like hotel bookings because they are trained on fixed layouts and stumble when a real website deviates from them. An agent that operates inside generated, model-understood interfaces would not face that brittleness. Runway explicitly frames Interface World Models as a training ground for agents.
The Catches: Text, Reliability and Accessibility
For now, significant problems keep Solaris firmly in the research column.
Stable, readable text remains a hard problem for generative video models — and interfaces are full of text. Buttons, prices, addresses and search fields all depend on exact strings, where a hallucinated character is a broken product.
Long sessions and reliability are open questions too. The Decoder notes a subtle but serious risk: a generated interface that is wrong but convincing-looking can be worse than no interface at all, because the user has no way to tell the difference between a rendered price and a real one.
Accessibility is another unsolved issue. A generated interface has no underlying structure for assistive technology to read, so screen readers — the primary way blind users navigate software — have nothing to work with.
Research For Now
Runway is not shipping Solaris as a product. The company is looking for partners for a launch and offers early access through a sign-up form, positioning the system as an exploration of what software could look like when interfaces are generated rather than programmed.
The timing is not accidental. World models have matured quickly as a category, with major labs racing to simulate realistic environments for robotics, gaming and video generation. Pointing that same machinery at software surfaces is a natural extension — and one that puts Runway, best known for video generation tools used by filmmakers, into direct conceptual competition with the productivity software establishment.
Whether anyone actually wants interfaces that keep changing is a question Runway leaves open. Software conventions — the shopping cart, the search bar, the settings gear — exist because they are predictable. An interface that reinvents itself for every user trades that predictability for adaptability, and it remains to be seen whether users experience that as magic or as chaos.
What is clear is the direction of travel. The same world models built to simulate physical reality for video are now being pointed at software reality — and Runway is betting that the next interface is not designed at all, but generated.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
