Speridlabs

Speridlabs is a spatial AI lab developing foundation models designed to understand and work with the 3D world. Its approach centers on a coherent spatial representation, where objects, surfaces, geometry, and occlusion remain consistent as a scene is viewed or navigated from different perspectives. The lab sees this shared foundation as a way to bring perception, reasoning, simulation, reconstruction, editing, and generation together rather than relying on separate models for narrowly defined tasks.

Speridlabs is developing a 3D world model intended to build coherent, navigable scenes and objects from text prompts or a small number of images. The model is listed as unreleased. Its published research includes Iris-3B, an open text-to-image model that generates pixels without a variational autoencoder, and ENEAS, a method for finding, segmenting, and tracking concepts across images and video using plain-text descriptions. Through its research, platform, and company updates, Speridlabs focuses on closing what it describes as the gap in foundation models for 3D understanding.