Video-As-Prompt and MotionGPT3 are published at ICLR 2026, exploring unified semantic video control and motion-language modeling.