Software That Generates Itself While You Use It
Runway just announced something that does not fit neatly into any existing category. Solaris, unveiled August 31, is what Runway calls an “Interface World Model.” Instead of rendering a website from code or displaying an app built from fixed components, Solaris generates the interface itself, frame by frame, as you interact with it.
Every screen you see is synthesized in real time. Click a shirt in a virtual store and drag it onto yourself. The interface responds. Move a table across a room. The lighting shifts. Change the colour of a couch. Done. There is no code underneath generating these responses. There is a world model predicting what the next frame should look like based on what you just did.
This is not a video editor. It is not a design tool. It is a new thing: an operating system layer where the interface is the output of a generative model, continuously rendered based on your input.
What Solaris Actually Does
Runway adapted its Gen-4.5 video model to understand interaction. When you click, drag, or type, Solaris treats those inputs the same way it treats text or image prompts: as conditioning signals for the next frame. A separate language model interprets your intent and decides what should happen. The world model renders the visual result.
The demos show three capabilities that are genuinely different from anything currently available:
- Entirely visual interfaces. No code layer sits beneath what you see. The image is the application. You can browse a virtual clothing store, pick up a shirt from a rack, drag it onto yourself to try it on, or rearrange the display. The interface is the experience.
- Living environments. Because the interface is continuously rendered, it is always evolving rather than waiting for the next user action. Reflections shift with the lighting. Objects respond naturally when manipulated. You can say “Move the table so I can see how it looks” or “Change the colour of the couch” and the scene adapts in real time.
- Open-ended interaction. Traditional interfaces can only do what developers anticipated during development. Solaris can support entirely different behaviours in the same scene, reacting to user interactions in real time. The capabilities of the interface are determined by the model, not by a predefined workflow.
Why This Is Not Just Another AI Demo
The “Interface World Model” framing matters because it addresses a specific bottleneck in how software gets built today. Right now, every piece of software requires a translation step: a visual design must be converted into code before it can do anything. That translation limits what an interface can be and how it responds to people.
Solaris removes that translation step. The visual design and the interactive behaviour are handled by the same model, jointly. No conversion. No intermediate representation. No loss between what was designed and what actually runs.
For small businesses and solo creators, this is significant because of what it eliminates: the need to code every interaction, build every page, and anticipate every user path before launch. The model handles rendering and response together.
The Three Big Engineering Problems Runway Claims to Have Solved
Building something like this required solving three problems that have kept real-time generated interfaces from working until now:
- Speed. Interactive interfaces stop feeling interactive around 500 milliseconds of delay. Video diffusion models take seconds or minutes to produce a clip. Solaris generates frames sequentially, with each frame depending only on what came before, cheaply enough to keep up with user input.
- Coherence. An interface has to stay consistent across an entire session. The things it needs to preserve, including text, layout, and object identity, are exactly the things generated video has historically struggled to maintain. Small errors compound the longer generation continues. Runway says it trained Solaris on its own distilled outputs to improve stability during longer interactions.
- Cost. Generating every frame is more expensive than serving a page that was built once. Runway claims Solaris is orders of magnitude cheaper to run than a standard video diffusion model, though it has not published pricing.
What to Look Out For
Solaris is a research preview, not a shipping product. Runway is accepting early-access requests and has not announced pricing, general availability, or API access. The demos look impressive, but there are known limitations: stable and legible text, extended-session coherence, and grounding responses in verified information all remain unsolved problems.
Accessibility is another open question. A synthesized frame does not automatically expose the document structure, labels, and controls that screen readers depend on. Runway says generated interfaces will need a separate accessibility layer, which does not exist yet.
Despite those caveats, the direction is clear. If Solaris or something like it reaches a usable state, it changes what a “website” even is. Instead of designing pages and coding interactions, you describe a scene and the model renders the experience. The interface is not assembled from components. It is generated as you go.
What This Unlocks for Creators and Small Businesses
- Product experiences without development. Imagine a virtual showroom where customers walk through a 3D space, pick up products, and see them from every angle. No Unity developer needed. No WebGL. The model generates the showroom as the customer explores it.
- Tutorial environments that respond to the learner. Instead of a fixed video walkthrough, a training experience that adapts in real time to what the learner does. Drop materials into a simulated experiment and see what happens. The model generates the outcome.
- Rapid prototyping at the experience level. Designers can sketch an interface concept and see it rendered as a living, responding experience in seconds. Not a static mockup. Not a coded prototype. The actual thing, generated on the fly.
- Agent training environments. Solaris can generate endlessly changing interface layouts and interactions. For training computer-use AI agents, this means exposure to layouts and workflows that never existed before, reducing fragility on unfamiliar sites.
Where It Stands Right Now
Solaris runs at 720p and targets interactive speed. Runway compared it against a coded interface generated by Claude Opus 5. In a study with 250 participants and 7,500 pairwise judgments, participants preferred Solaris for following instructions 61% to 24% and for behaving naturally 71% to 21%. Those are strong numbers, but they measure visual fidelity and natural feel, not reliability, accuracy, or production readiness.
The benchmark says Solaris is better at making an interface feel alive. It does not tell us whether it can preserve an account balance, submit a tax return once and only once, or remember that “delete” was supposed to require confirmation.
This is where things actually stand: Solaris is a convincing demo of a new category. It shows that generating interfaces in real time is possible at interactive speed. The problems that need solving before it replaces your website are real, but so is the signal that the interface layer of software is about to become generative.
Watch this space. The shift from coded interfaces to generated ones will not happen overnight, but the starting gun just fired.
Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.
