Sand.ai is an artificial intelligence company focused on advancing AI technology and making its benefits accessible to everyone. Its work centers on generative media, with MAGI-2 Preview as its featured model. MAGI-2 is a unified audio-video generation system with 114 billion parameters, activating 6 billion parameters per token. Built on the company’s MagiMoE approach and developed across architecture, systems, and data, it explores a more efficient way to scale video generation.

The model is designed to create expressive character performances, including dialogue and singing, with coordinated lip-sync, facial expressions, eye movements, and body language. It generates audio and visuals jointly to align speech with lip movements, actions with sound effects, and environmental changes with audio. MAGI-2 also responds to cinematographic direction, including shot scale, composition, perspective, and camera movements such as zooming, tracking, and orbiting.

Sand.ai provides an API platform for accessing its technology and showcases creative partner projects across advertisements, music videos, animation, and other visual formats. Its stated mission is to advance AI for everyone.