Starchild-1

#243 en Modèles de langage Frontier

Odyssey · v1 · depuis 2026-05-17 · 13× · vu le 30 juin 2026

3

Momentum

Starchild-1 is the first real-time multimodal world model from Odyssey (odyssey.ml), autoregressively generating synchronized audio and video in real time while continuously responding to streamed user inputs (text, speech, actions). Released as a preview with no public interactive demo yet available. Technically, it relies on a causal distillation pipeline that converts a bidirectional audio-video foundation model (Ovi) into a real-time autoregressive world model, plus an asynchronous KV-cache architecture handling the differing temporal frequencies of audio and video. Target domains include gaming, robotics, education, healthcare, and defense.

Preuves (13)