Top AI Product

Odyssey ML announced Starchild-1 on May 17 — the first general world model that autoregressively generates synchronized audio and video in real-time while continuously responding to streaming user input. The kicker: world models until now have been silent.

## What’s actually new

Previous world models (Genie, Sora video, Decart’s models) learned visual dynamics from large-scale video but couldn’t generate audio. Starchild-1 generates synchronized audio and video, autoregressively, in real-time, as the user keeps streaming input. The result is an interactive world simulator you can hear walking through, not just watch.

## How it learns

World models train directly from raw pixels, motion, and actions encoded in massive video corpora — no language layer, no text supervision. Adding audio means the model is now learning that footsteps land when feet touch the ground, that voices come from mouths, that drums match drumsticks. Multimodal grounding without the language detour.

## Why it matters

World models are widely viewed as the path to embodied AI, robotics simulation, and game-engine-grade interactive media. Starchild-1 is the first to cross the audio-video sync threshold. Compared with LLMs we’re still in “GPT-2 era” for world models — but the shape of the next decade is now visible: language models for thinking, world models for acting in space and time.