MotionHiFlow: Text-to-motion via hierarchical flow matching

ArXi:2604.23264v1 Announce Type: new Text-to-motion generation aims to generate 3D human motions that are tightly aligned with the input text while remaining physically plausible and rich in fine-grained detail. Although recent approaches can produce complex and natural movements, they usually operate at only one temporal scale, which limits both semantic alignment and temporal coherence.