Runway is pushing generative AI beyond images and videos with a new experimental system called Solaris, which can generate interactive software interfaces in real time.
Unlike traditional AI coding tools that generate HTML, CSS, JavaScript or other programming code, Solaris takes a radically different approach: it generates the interface itself, frame by frame, as the user interacts with it.
Runway calls Solaris its first Interface World Model, a new category of AI system designed to make software behave more like an interactive environment than a collection of predefined screens.
The technology is still an early research project rather than a finished commercial product. Runway says it is working with partners toward a public launch and is currently offering early access.
What Is Runway Solaris?
Solaris is an AI system designed to generate interactive interfaces continuously.
Today’s software generally follows a predictable development pipeline. A designer creates a visual concept, developers translate that design into code, and the resulting application runs using technologies such as HTML, CSS, JavaScript or native software frameworks.
Solaris attempts to remove that intermediate layer.
Instead of generating code that eventually produces an interface, Solaris generates the visual interface directly.
Every frame is synthesized as the user interacts with it, meaning a click, drag or other action can influence what appears in the next frame.
Runway describes this as eliminating the traditional intermediate representation between an interface concept and the final experience.
Solaris Doesn’t Build Apps the Traditional Way
The most important distinction is that Solaris is not simply another AI coding assistant.
An AI coding tool might receive a prompt such as:
Build an online clothing store.
It could then generate React, HTML, CSS and JavaScript code that developers can modify and deploy.
Solaris approaches the problem differently.
The visual environment itself becomes the application.
Runway’s example is a virtual clothing store where a user could interact directly with products. Instead of navigating through conventional product pages, a person could pick up a shirt in a generated showroom and drag it onto an image of themselves to see how it looks.
The same concept could apply to other interfaces.
A room-design experience could allow a user to tell the system to move a table or change the color of a couch. A cooking environment could let users drag ingredients into a bowl and see the scene respond.
The interaction isn’t necessarily limited to buttons and menus that developers previously programmed.
How Solaris Generates an Interface
Solaris builds on Runway’s Gen-4.5 video-generation model and the company’s earlier GWM-1 world-model research.
Runway adapted its video-generation technology so that it could understand interaction and generate responses continuously.
The process has several important components.
User Actions Become Model Inputs
When a user clicks, drags or modifies something, that action becomes conditioning information for the next generated frame.
Instead of a programmer writing:
When button X is clicked → execute function Y
the model learns relationships between actions and visual outcomes.
This allows the same interface to potentially respond to interactions that were not explicitly designed as fixed workflows.
Frames Are Generated Sequentially
Traditional video-generation models often generate clips through multiple denoising steps, which can take too long for interactive software.
Runway says it converted Solaris into a real-time engine using three stages.
First, it trained the system to generate frames autoregressively, with each frame depending on what came before.
Second, it distilled the multi-step generation process into fewer steps.
Third, Runway trained the faster model on its own outputs to help preserve visual quality during longer interactions.
The result is intended to operate at interactive speeds while maintaining visual consistency at 720p.
A Language Model Handles the Reasoning
Solaris isn’t operating entirely by itself.
Runway describes a system in which a language model handles the reasoning while Solaris handles visual rendering.
The language model interprets what the user wants, determines how an interaction should affect the current scene and generates guidance for the visual model.
Solaris then generates the resulting interface frame.
In simplified terms:
User interaction → Language model reasoning → Solaris rendering → New interface frame
That separation is important because it allows the system to combine natural-language understanding with continuous visual generation.
Why Runway Thinks This Could Change Software
The most interesting part of Solaris isn’t simply that it can generate attractive interfaces.
It is the possibility of making software open-ended.
Traditional applications are limited by what developers anticipate during development.
A shopping website might have product pages, filters, a cart and checkout flow because those interactions were programmed.
Solaris could potentially allow the interface itself to change based on what the user is trying to accomplish.
Runway argues that this could make the interface less like a fixed collection of pages and more like a continuously generated environment.
That could eventually lead to software that creates a customized interface for each person and each task.
Solaris Could Make Apps Feel More Like Virtual Worlds
One of Runway’s biggest ideas is that interfaces could become more visual and spatial.
Imagine opening a furniture-shopping experience.
Instead of browsing hundreds of product pages, you might enter a virtual showroom. You could move furniture around, change materials and experiment with different layouts directly within the environment.
The interface would not simply show buttons labelled “Change Color” or “Move Object.”
The objects themselves could become the controls.
Runway describes this as making software feel more like interacting with a living environment than navigating through scripted pages.
Solaris Could Also Help Train AI Agents
There is another potential application that could be even more important.
AI agents increasingly need to operate computers on behalf of users.
The problem is that traditional software interfaces are often rigid. An agent may learn how to interact with one website but struggle when another website uses a different layout.
Solaris could provide AI agents with continuously changing interfaces.
Because the environment can be generated dynamically, agents could potentially train against interfaces they have never encountered before.
Runway specifically highlights this as a potential use for Interface World Models.
That means Solaris may eventually be useful not only for human-facing applications but also as an environment in which AI agents learn how to interact with software.
Runway’s Early User Study
Runway has also compared Solaris-generated interfaces with coded interfaces.
In a study involving 250 participants and 30 interaction examples, Runway reported that participants preferred Solaris-generated interfaces 61% of the time for following the requested interaction and 71% of the time for natural behavior within the scene. The comparison was made against coded interfaces generated using Claude Opus 5.
These figures are interesting, but they should be treated as company-reported results, rather than independent evidence that Solaris is universally better than conventional software.
The technology is still experimental, and independent validation will be important as the system becomes available more broadly.
Solaris Has Some Major Problems to Solve
The concept is impressive, but generated software also introduces problems that traditional software largely solved years ago.
Text Rendering
Readable and stable text remains difficult for generative visual models.
That is particularly problematic for software because interfaces often contain large amounts of precise text, numbers, labels and data.
A slightly distorted word in a generated video might be harmless.
A distorted word in a banking application could make the software unusable.
Runway acknowledges stable text as one of the major challenges.
Accessibility
Traditional websites and applications can expose structured information to screen readers and accessibility systems.
A generated visual interface does not automatically provide the same semantic structure.
Runway identifies accessibility and integration with technologies such as screen readers and accessibility APIs as an important challenge.
Accuracy and Trust
Another issue is that a generated interface can look convincing even when something is wrong.
For applications involving commerce, education, finance, healthcare or other sensitive areas, visual realism is not enough.
The system must also be grounded in reliable information.
Runway says it is exploring ways to ground generated experiences using product imagery and reference material.
Long Sessions
Keeping an interface coherent for a few seconds is easier than maintaining consistency during a long session.
Objects, text, layouts and visual details can gradually drift in generative systems.
Runway identifies long-term coherence as one of the key engineering challenges for Solaris.
Cost
Generating every interface frame dynamically is computationally more expensive than serving a conventional webpage that was built once.
Runway says its real-time architecture makes Solaris significantly cheaper to run than standard video diffusion, but dynamically generating software still has a fundamentally different cost profile from serving conventional code.
Is Solaris Available to Everyone?
No.
Solaris is currently an early research system.
Runway says it is working with key partners toward a public launch and is accepting requests for early access. There is currently no broadly available consumer product or publicly announced general pricing.
That distinction is important.
Solaris demonstrates where Runway believes software interfaces could go, but it does not mean developers can currently replace their entire production stack with Solaris.
Does Solaris Mean Traditional Apps Are Going Away?
Not anytime soon.
Conventional software has major advantages: speed, reliability, accessibility, security, predictable behavior and mature development tools.
For many applications, deterministic code remains much better suited to production.
Solaris is instead exploring a different model in which an interface can be generated dynamically around a user’s intent.
If the technology becomes reliable enough, the distinction between “opening an app” and “asking an operating system to create an experience” could eventually become less important.
The Bigger Idea Is AI-Generated Software
Runway’s Solaris may be more significant as a new software architecture experiment than as an immediate replacement for app development.
For years, generative AI has been moving down the software stack.
First came AI assistants that generated code.
Then came systems that could generate entire applications from natural-language prompts.
Solaris proposes something different: remove the code-generated interface from the interaction loop entirely.
That is a much more radical idea.
If it works at scale, an operating system could potentially generate a custom interface whenever a user needs one.
Need to compare products? Generate a visual shopping environment.
Need to learn chemistry? Generate an interactive experiment.
Need to redesign a room? Generate a spatial design interface.
Need to understand a complicated machine? Generate an interactive simulation.
The app may no longer be the primary unit of software.
The task could become the primary unit.
However, there is a long distance between an impressive research demonstration and reliable production software. Solaris still needs to solve text accuracy, accessibility, grounding, consistency, security, cost and integration.
For now, Solaris is best viewed as a glimpse into what Runway believes the next generation of software interfaces could look like.
What Happens Next?
Runway says it is working with partners toward a public launch and continuing research into richer interactions, stronger grounding and longer-lived generated experiences.
The biggest question is whether Interface World Models can move beyond demonstrations and become reliable enough for everyday software.
If they can, the impact could extend far beyond AI-generated websites.
It could change the basic assumption that software must be designed first, coded second and then used by everyone through the same predefined interface.
Solaris is betting on a different future: software that generates itself around the person using it.
Read More:- Google Expands Gemini Across Android as Google Assistant Shutdown Nears
Frequently Asked Questions
What is Runway Solaris?
Runway Solaris is an experimental AI system that Runway calls an “Interface World Model.” It generates interactive software interfaces frame by frame in real time instead of relying on traditional HTML, CSS and JavaScript.
Can Solaris generate complete apps without coding?
Solaris is designed to generate the interactive interface without a traditional code layer underneath. However, it is currently a research system and should not be confused with a fully available production-ready app builder.
How does Solaris work?
Solaris uses a modified version of Runway’s Gen-4.5 video-generation technology. User actions such as clicks and drags become inputs for the next generated frame, while a language model helps determine how the interface should respond.
Is Runway Solaris publicly available?
Not yet as a general consumer product. Runway is working with partners toward a public launch and currently offers an early-access process.
What are the biggest limitations of Solaris?
Major challenges include stable text rendering, accessibility, accuracy and grounding, long-session consistency and the cost of generating interfaces continuously.
Could Solaris replace traditional apps?
It is too early to say. Solaris could eventually enable dynamically generated interfaces, but traditional applications remain much more mature for reliability, accessibility, security and predictable behavior.
Could Solaris be useful for AI agents?
Yes. Runway specifically sees dynamically generated interfaces as a potential training environment for AI agents, allowing them to learn to interact with changing interfaces rather than memorizing fixed layouts.
