Naver's "Seoul World Model" uses actual Street View data to stop AI from hallucinating entire cities

THE DECODER / 3/29/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

Key Points

  • Naver developed the “Seoul World Model,” a video world model grounded in real city geometry derived from over one million Street View images.
  • By grounding generation in actual spatial structure, the approach is designed to reduce the tendency of AI systems to hallucinate entire cities.
  • The model can generalize to other cities without requiring fine-tuning, suggesting the learned representations transfer across environments.
  • The work highlights a shift toward reality-grounded generative modeling that uses large-scale geospatial data to improve fidelity and reliability.

Neon illustration of Godzilla appearing between pink and blue skyscrapers in a busy street in Seoul.

South Korean internet giant Naver built a video world model grounded in actual city geometry from over a million of its own Street View images. The model generalizes to other cities without any fine-tuning.

The article Naver's "Seoul World Model" uses actual Street View data to stop AI from hallucinating entire cities appeared first on The Decoder.