English

Model ReleasesWorld LabsAtlas

World Labs Announces World Model Atlas, Enabling 3D Space Generation from Few Images and 1440p Video Output

This article is a translation. Read the Japanese original

AI company World Labs has announced "Atlas," a next-generation world model that integrally handles text, images, and 3D space. A world model refers to an AI model capable of learning and reproducing the physical properties and object structures of the real world.

Atlas maintains 3D spatial context and supports generation via free camera control as well as spatial reconstruction from a small number of images. A key feature is its ability to generate a 3D space from a single image, which is said to enable camera work that remains faithful to instructions. Furthermore, it is reported to be capable of video generation at resolutions up to 1440p, and can be utilized for creating bullet-time footage and simulations for robotics.

The company is currently providing early access to select partners and is reported to plan a sequential integration into its own products, such as "Marble."

Sources

  1. フェイフェイ・リー氏のWorld Labs、空間知能オムニモデル「Atlas」発表 少数の画像から3D空間やバレットタイム映像を生成 (ITmedia AI+)
  2. 画像も動画も3D空間も生成可能な世界モデル「Atlas」が登場、「1枚の画像から3D空間を生成」「指示に忠実なカメラワーク」「1440pの動画生成」など多機能なモデル (GIGAZINE)