Runway Solaris Generates Interfaces Without Coded Apps
The research system renders interactive 720p interfaces frame by frame, extending generative video into software and agent training.
Software rendered as video
Runway has introduced Solaris, an early research system that generates an interactive visual interface continuously instead of executing a conventionally coded application. The company calls it its first “Interface World Model,” positioning the technology as a bridge between generative video, software interfaces and environments for training computer-using agents.
Solaris begins from an image or described scene and synthesizes the interface one frame at a time at up to 720p. Clicks, drags and typed instructions condition subsequent frames, allowing objects and layouts to respond without developers defining every interaction in advance. A language model interprets the user’s intent and decides how the scene should change, while a world model derived from Runway’s Gen-4.5 technology handles visual rendering.
Runway says it converted the underlying diffusion system to autoregressive generation, reduced the number of denoising steps through distillation and trained the faster model on its own outputs to improve stability during longer sessions. In a company-run study involving 250 participants and almost 7,500 pairwise judgments, Solaris was preferred over a Claude Opus 5-generated coded interface in 61% of instruction-following comparisons and 71% of natural-behavior comparisons.
Solaris is not yet a public product; Runway is recruiting early-access partners. The company also acknowledges unresolved problems involving text precision, deterministic behavior, accessibility, security, integration with conventional software and the cost of generating every frame.
Why it matters
Solaris expands world models beyond video clips and simulated physical spaces into the software layer itself. Constantly changing interfaces could provide richer training environments for agents that currently overfit to familiar website layouts. They could also produce highly personalized visual applications that would be expensive to hand-code.
Yet generated pixels do not automatically supply the reliability, auditability or accessibility expected from production software. Solaris is therefore best understood as a credible research direction—and a strategic expansion for Runway beyond media generation—rather than an imminent replacement for coded applications.