Runway Debuts Solaris to Render App UIs as Live Video

Runway has introduced Solaris, its first Interface World Model designed to generate application user interfaces frame by frame in real time at 720p resolution. Built on the company's Gen-4.5 and GWM-1 architectures, Solaris separates the reasoning process, handled by a large language model, from the visual rendering, managed by the world model. Instead of compiling design mockups into traditional code like HTML or JavaScript, the system synthesizes pixels dynamically in response to user actions like clicks and drags. To achieve interactive speeds, the model uses autoregressive generation, step distillation, and self-training.

In evaluation benchmarks, Solaris outperformed GPT-4o, Gemini 2.5 Pro, and Fable 5 on interface reconstruction tasks, which were measured using SSIM and DINOv3 similarity metrics across 30 distinct interfaces. Additionally, a user study featuring 250 participants making nearly 7,500 pairwise judgments compared Solaris to Claude Opus 5. Evaluators preferred the Solaris-generated video over the coded alternative in 61% of instruction-following comparisons, compared to 24% for the coded results. For natural behavior within the scene, Solaris was preferred 71% to 21%.

For developers and designers, this technology could eliminate the lossy translation step between visual mockups and frontend code, allowing the operating system to generate custom interfaces on demand. It also offers a novel training environment for AI agents, which often struggle to generalize across static layouts. However, Runway acknowledges several current limitations that practitioners must consider before production use. These include difficulties with stable text rendering, maintaining long-term session coherence, ensuring factual grounding, and integrating with accessibility APIs. Solaris is currently not publicly available, though Runway is accepting early access requests.