Meta AI's VGGT Model Brings Fast 3D Scene Reconstruction to Replicate Platform

A developer known as Vufinder has deployed VGGT-1b-Point, a 3D vision AI model originally developed by Meta AI Research and the University of Oxford, on the Replicate platform. The model, which won Best Paper at CVPR 2025, uses a single feed-forward neural network pass to extract camera parameters, depth maps, and 3D point clouds from one or more images in under a second. Unlike traditional 3D reconstruction pipelines that require hours of geometric optimization, VGGT delivers state-of-the-art results without post-processing steps. It accepts common image formats and video files, processing inputs up to 518 pixels in maximum dimension. Use cases include robotics spatial mapping, video camera calibration, 3D point tracking, and novel view synthesis.