Runway has taken its video generation technology and aimed it at something far more ambitious than producing polished clips. Solaris, the company's newest model, generates fully interactive interfaces in real time, frame by frame, with no underlying code required.

Announced on August 31, 2026, Solaris is the first entry in what Runway calls its Interface World Models family. It is less a chatbot that writes code for you and more an AI that is the application—rendering every visual element and responding to every user input simultaneously through a single model.

What Solaris Actually Does

Traditional software separates visuals from logic. Designers create the look, developers wire up the behavior, and an entire pipeline of rendering engines, frameworks, and interaction handlers stitches it all together. Solaris collapses that entire stack into one system.

The model generates each frame as a direct response to what the user does. Click a button, and Solaris does not execute pre-written code to display the next screen. Instead, it synthesizes a new visual frame reflecting the interaction, maintaining coherence across the full session at 720p resolution.

In practical terms, the generated image is the application. There is no DOM, no CSS, no JavaScript underneath. The AI handles rendering and interaction jointly, meaning there are no intermediate representations to slow things down or introduce bugs.

Runway built Solaris on top of its Gen-4.5 video generation model, the same foundation that powers its video creation tools. Video models already understand temporal coherence—how one frame should logically follow another. Solaris adds a feedback loop: instead of following a text prompt's trajectory, it follows the user's actions.

For non-programmers, the appeal is straightforward: if the interface itself is generated on demand, ideas for simple tools or experiences could become usable without writing or inspecting any code at all. The approach also carries open questions of its own—for instance, how reliably a purely generated interface can maintain exact text, numbers, or layout details across a long session without the determinism that traditional rendering guarantees.

From Video Clips to World Models

Runway, founded around 2018, made its name with generative video tools that let creators produce cinematic footage from text descriptions.

In December 2025, Runway announced GWM-1, its General World Models initiative, focused on autoregressive models designed for applications well beyond video. That announcement laid the intellectual groundwork for what Solaris represents: AI systems that do not just generate media but understand and simulate environments.

The collaboration with NVIDIA has also been a key ingredient. Real-time inference at the quality Solaris promises requires substantial hardware capacity, and Runway has been working with NVIDIA specifically on the compute infrastructure needed to make frame-by-frame generation fast enough for interactive use.

The Competitive Landscape

Solaris enters a market where the boundaries between AI tools and traditional software are already blurring quickly. Companies like Vercel, Replit, and Cursor have been racing to let AI generate functional applications from natural language prompts. Those approaches, however, still produce code as an intermediary.

No independent benchmarks existed at the time of the announcement. Runway's claim that Solaris surpasses other language models in performance has not been verified by third parties.

The 720p resolution ceiling also raises questions. For simple interactive experiences, prototypes, or mobile-first applications, it is workable. For desktop productivity software or data-dense dashboards, it could feel limiting.

What will indicate whether Solaris is more than a research demo is how the next months unfold: third-party testing of its consistency and latency, whether Runway raises the resolution ceiling, and whether developers and designers adopt it for real prototyping workflows or treat it as a novelty.