Per-Stroke Temporal Control for Text-to-Motion via Action Units and Action-Detection Guidance

A new text-to-motion method adds per-stroke timing control, aiming to make prompted actions land on the right frames instead of just the right motion. It introduces Action Units that encode body track, action class, time window, and impact timing, then uses a lightweight adapter plus detector-guided refinement to tighten placement. For animation and tools teams, it pushes motion generation closer to usable production control.