Motion Synthesis Models Still Demand "A person..." Prompts — Just Rewrite the Captions

andrew_n_carr · x · 2026-11-17

Andrew Carr's pet peeve with the motion synthesis community: because everyone trains on HumanML3D or AMASS, all models require prompts starting with "A person ...." His fix: it's 2026 — have your agents rewrite the captions before training to break free of this dataset legacy.

Original post →

More from Research

Research channel →