World Labs co-founder Justin Johnson describes Atlas as a base model for generating environments, reconstructing real spaces from images and simulating how systems behave inside them. The work is intended to support applications across visual effects, games, virtual reality, construction and robotics.
Johnson says Atlas grounds reference images in three-dimensional space and lets users steer exact camera paths. Its architecture can produce direct two-dimensional frames or explicit three-dimensional representations, allowing applications to choose between model-generated pixels and assets that fit established production pipelines.
Johnson positions world models as a general platform that complements language models. For robotics, he describes reconstructing a workplace from phone images, then using that simulation for evaluation or adaptation before deployment in the real environment.
Watch on YouTube
