Sekai World Video Dataset

Sekai is a high-quality first-person perspective global video dataset released in 2025 by Shanghai Artificial Intelligence Laboratory and Shanghai Creative Institute. The related paper results are:Sekai: A Video Dataset towards World Exploration", aims to inspire valuable applications in the fields of video generation and world exploration.

This dataset focuses on ego-centered world exploration and consists of two parts: Sekai-Real and Sekai-Game. It contains more than 5,000 hours of walking or drone perspective (FPV and UVA) videos from more than 100 countries and regions and 750 cities. The videos are long (1 to 39 minutes, 2 minutes on average) and contain rich exploratory annotations and world annotations. Each video is annotated with a variety of labels, including subtitles, location, scene, weather, crowd density, time, and camera trajectory.

Dataset Overview