Runway's Solaris Wants to Skip Code and Generate Software Frame by Frame

Runway spent years teaching the world what "generative video" meant, first as a tool for filmmakers, later as a broader pitch about AI and visual media. At its AI Summit in San Francisco this month, the company stepped through a new door. It introduced Solaris, the first product in a family it is calling Interface World Models, and the framing is as ambitious as the name.

The idea is this. Today, when you design software, you draw it, then someone compiles those designs into code, and the code renders pixels. Runway's argument is that the compiling step is a lossy compression of what the software could have been. By the time a user arrives, the interface is frozen. Solaris removes that middle step. Instead of turning a design into code, it renders the interface itself, frame by frame, in real time, generating the visuals and the response to input together.

Runway's CTO Kamil Sindi put it plainly at the summit. Today's software is a lossy compression of possible interactions. Solaris, in this telling, is a self-building operating system where software is generated without a single line of code.

It is a sharp pivot for a company that most people still file under "video." The summit also brought Runway Ads, an autonomous performance-marketing engine claiming a 1,000 percent increase in ad volume with doubled return on ad spend, plus a broader repositioning toward agent-ready infrastructure. Co-founder Anastasis Germanidis told the room that universal world simulators will be the most important technology of this era. That is not the language of a video editing vendor.

There is real momentum behind the shift. Runway has been valued around $5.3 billion, and earlier reporting described a $10 million venture fund and a Builders Program meant to formalize relationships with studios and developers building on its APIs. The company has talked for a while about reaching "video intelligence" through video, and Solaris is the first product that tries to make that concrete.

The technical bet is genuinely interesting, and also genuinely risky. Rendering an interface frame by frame means every pixel is recomputed every time, which is expensive and can feel slow or flickery compared to the declarative rendering the web has spent twenty years perfecting. A world where interfaces are generated rather than coded raises hard questions about determinism, testing, and what happens when the same input produces a slightly different layout on the next frame. Code is not just a cost to be removed. It is also how software becomes predictable, auditable, and reproducible.

The timing is not accidental. Runway made this pivot the same week OpenAI was winding down its consumer video products and redirecting the Sora team to world simulation. The word "world model" has become the term everyone reaches for when they want to signal that generative media is a stepping stone to something bigger than pretty clips. Runway is betting that the stepping stone leads to software that builds itself, and it is making that bet loudly enough to attract a venture fund and a developer program around it.

There is a business logic underneath the vision. A video generation tool sells to creative teams, a bounded market. An infrastructure layer that any agent can call to generate and reason about visual media sells to everyone building on agents, which is a much bigger addressable market. Runway's repositioning from tool to platform is the classic enterprise play: stop selling the product, start selling the thing everyone builds products on.

Whether Solaris becomes a real product category or a memorable demo depends on the boring parts: latency, cost, and whether a generated interface stays consistent enough to actually use. Those are the same questions that have humbled every "no-code" promise before it. The history of no-code is a history of demos that break on the second click, and a frame-by-frame renderer inherits all of those risks at a higher compute cost.

Still, the direction Runway is pointing at matches a broader current in the industry, and it is worth separating the vision from the specific product. The claim that interfaces could be generated rather than compiled is a bet on a future where the user, not the designer, is the one who shapes the software. That is a genuinely different idea about what an application is, and it has implications far beyond Runway.

It is worth pausing on what "world model" actually means in practice, because the term gets thrown around loosely. In Runway's usage, a world model is a system that understands enough about how things look and behave that it can predict and render the next state, the next frame, the next response. That is a genuinely different capability from a video generator, which produces a plausible clip without any understanding of what it is depicting. The bet is that the same model that can render a character walking can also render a button responding to a click, because both are just "the world reacting to input." Whether that bet pays off is an open question, but it is a coherent one, and it is why Runway keeps saying it is not a video company anymore.

The contrast with OpenAI is instructive. OpenAI shut down its consumer video products and pointed its Sora team at world simulation for robotics, an internal research direction with no commercial product attached. Runway is trying to do world simulation as a product, with a venture fund and a developer program around it. Two companies arrived at the same conclusion about where the technology is heading and took opposite paths on how to pursue it.

There is also a practical audience question hanging over the whole thing. Interface World Models, if they work, are aimed at people who want software without writing code, which is a huge and famously hard market. Every generation of "no-code" promised this and delivered tools that work for demos and break for real products. Runway's version is more radical because it removes the intermediate representation entirely, but the same gap applies: a demo that handles one happy path is not the same as a product that handles the edge cases.

One more thing worth watching is how this changes the economics of the design-to-code pipeline, an industry that has absorbed a huge amount of venture money. If an interface can be generated and rendered directly, the entire category of tools that turn Figma files into frontend code faces an existential question. Runway is not just pitching a new product; it is pitching a future where that translation layer does not exist because there is nothing to translate.

For developers and designers, the takeaway is not to panic or to celebrate. It is to watch. If Interface World Models work, they change how a lot of software gets made, and the skills that matter shift from writing layout code to describing intent. If they do not, Runway still learned something the rest of the field will be studying for years. Either way, the video company has stopped talking only about video, and that is the real announcement. The company that defined generative video is now trying to define what comes after it, and everyone who builds software should be paying attention to whether it succeeds.