Runway's Solaris generates app interfaces in real time
Instead of running code, the system renders each frame of the interface at 720p. Runway calls the new category an Interface World Model.
Symbolic image: a hand drags a glowing interface panel across a large touchscreen while the status lights of a compute workstation blink beside it.
Runway has unveiled Solaris, an AI system that does not ship software interfaces as code but instead draws them frame by frame at 720p while a person is using them.
At a glance
- Runway calls Solaris the first model in a category it names Interface World Models.
- The interface is produced frame by frame at 720p; clicks, drags and voice commands feed into the next frame.
- It builds on Runway's Gen-4.5 video model and continues the approach begun with GWM-1.
- A language model decides what changes on screen; the world model renders the frame.
- Not a product: Solaris stays research, with early access handled through a partner form.
Runway has introduced Solaris, a system that stops handing a user interface over as executable code and instead paints it frame by frame while someone is operating it. Output runs at 720p. Clicks, drags and spoken commands become inputs to whatever the next frame shows. Runway labels the category an Interface World Model and presents Solaris as its first entry.
A language model decides, a world model draws
Solaris sits on top of Gen-4.5, Runway's video model. The split is two-stage: a language model works out what should change on screen, and the world model renders that decision as an individual frame. The approach continues the line Runway started with GWM-1.
The distinction is not only technical. A conventional interface exists before you open it; this one comes into being at the moment of use — and may come out differently the next time.
The uses Runway points to
- Online retail: shoppers drag a garment onto their own photo rather than reading one product page built for everyone.
- Product visualization: spoken instructions such as move the table or change the color of the couch alter the scene straight away.
- Tutorials: instead of a block of chatbot text, the system demonstrates the step inside the user's own context.
- Training AI agents: agents practice against layouts that shift rather than one frozen page.
The larger claim
Runway draws a bigger argument from this: the app as a fixed unit goes away. An operating system would then hold no catalog of pre-built programs and would generate the interface that fits the moment. That is the vendor's positioning, not a demonstrated finding.
What does not work yet
Stable, legible text is still a problem — awkward in a medium largely made of labels. Screen reader support is absent, which leaves accessibility unresolved. Reliability across longer sessions is untested. And the system can produce results that look convincing while being wrong.
The reporting gives no figures for latency, frame rate or compute cost, and no independent measurements. What is described here is the state of things as Runway presents it.
Availability
Solaris remains a research effort rather than a product. Runway is looking for partners and grants early access through a form. No general availability date and no price have been named.
FAQ
What is an Interface World Model?
A model that does not execute an interface as code but generates every frame of it, reacting to clicks, drags and voice commands as it goes. Runway presents Solaris as the first model of this kind.
Can I use Solaris yet?
No. By Runway's account Solaris is a research system. The company is seeking partners and offering early access via a form; no launch date or price has been given.
Which model is Solaris built on?
On Gen-4.5, Runway's video model. A language model decides how the interface changes and the world model renders the frames at 720p. The approach follows on from GWM-1.
