Starchild-1 · 项目原始材料
核对原文摘录 · 3 段，支持 6 项资料
原阅读方向 / 一句话认识 / 输出什么
autoregressively generates synchronized video and audio while responding to streaming open-ended text input.输入什么 / 动作是否影响预测
an action at timestep at to span visual control signals, speech content and auxiliary audio conditioning such as sound or music descriptions all of which we define in text space.材料说明的限制
extended interactive sessions can still display gradual drift in scene identity and acoustic environment over long horizons