Language-Native Control: Composes character and camera actions into textual instructions and injects them through MiniMax-H3’s pretrained text pathway. Temporally Grounded: Assigns one action prompt to each video latent interval, enabling precise control when actions change over time. Efficient & Ge
