On August 31, 2026, Runway announced Solaris, the first model in a new family it calls Interface World Models. Solaris generates interactive interfaces frame by frame in real time, eliminating the traditional code layer that sits between a design and a running application. A paired language model decides what the interface should do next; the world model renders the result at 720p resolution with a reported 37 ms per frame latency.

The company says Solaris is built on its Gen-4.5 video generation model, adapted to understand interaction and respond in real time through autoregressive frame generation, multi-step denoising distillation, and training the fast model on its own outputs for long-session coherence. Runway is not releasing Solaris as a public product; it is working with key partners toward a public launch and collecting early-access requests.

What's new

  • Entirely visual: The generated image is the application — no DOM, CSS, or JavaScript underneath. Users can drag a shirt onto a photo of themselves in a virtual store, or speak a command such as "Move the table so I can see how it looks" and watch the scene respond continuously.
  • Alive and continuous: Reflections shift with lighting, objects react naturally during manipulation, and the same starting frame can respond to the same drag in different ways depending on context.
  • Open-ended interaction: Interfaces are not limited to pre-scripted workflows. Clicking an object can redefine subsequent clicks — for example, clicking a cat lets subsequent clicks apply its fur texture to other objects.
  • Architecture: An LLM interprets user intent and guides scene evolution while the world model handles real-time rendering, separating reasoning from rendering.
  • Benchmarks: In a Runway-run preference study with 250 participants across 30 interaction examples (nearly 7,500 pairwise judgments), Solaris was preferred 61% of the time on instruction following (versus 24% for coded interfaces, 13% equivalent) and 71% on natural behavior (versus 21%). AlphaSignal reports Solaris beats GPT-4o, Gemini 2.5 Pro, and Claude Opus 5 on interface reconstruction, though Runway has released no independent third-party verification.

Why it matters

Solaris challenges the foundational split between "systems that know things" (LLMs, search) and "systems that respond in real time" (game engines, front-end code). By collapsing rendering and interaction into a single world model, it removes the translation step that historically loses visual fidelity and constrains possible interactions to what developers anticipated. The near-term competitive window, as multiple outlets note, narrows to rapid prototyping and synthetic agent-training pipelines where determinism and state guarantees are less critical than visual flexibility.

Our take

The elimination of the code layer is the headline, but the determinism gap is the constraint. Generated UIs cannot guarantee button identity, transaction integrity, or accessibility compliance — problems that coded interfaces solve by design. Until those gaps close, Solaris's strongest leverage is as a training substrate for agents that must operate in constantly changing visual environments, not as a drop-in replacement for production front ends.

Sources