MixiMotion generates human motions from text in one fast step
MixiMotion: One-Step Text-to-Motion Generation via Asymmetric Set Distillation
Computer Vision and Pattern RecognitionMultimedia
Summary
Creating animations of people moving based on written descriptions usually takes many steps and a lot of computing time. The paper introduces MixiMotion, a technique that can produce entire motion sequences from text with just one quick pass through a computer model. The method learns from many example motions and smartly matches these with generated ones to cover variety while avoiding errors. This approach achieves good quality nearly matching slower, multi-step methods but is around 89 times faster, making it useful for real-time applications.
What this means in practice
- •For game developers: Generate character animations from text commands instantly to enhance interactive gameplay experiences.$Commercial implications: Enables real-time text-driven animation for games, improving content creation speed and flexibility.
- •For virtual reality creators: Create diverse and semantically accurate human motions from textual descriptions to enrich immersive VR environments.
Authors
Hung Dinh, Binh Mai, Tran Quoc Bao Le, Lam Nguyen, Cong Tran
Abstract
Iterative text-to-motion generation delivers high-quality and semantically aligned motions but requires multiple network evaluations, resulting in substantial inference latency. We present \textbf{MixiMotion}, a strict one-step text-to-motion generation framework based on offline set distillation. Instead of distilling a single teacher trajectory for each text prompt, MixiMotion constructs an offline bank of multiple teacher motions and aligns teacher and student sample sets through \textbf{asymmetric bidirectional matching}. The teacher-to-student direction promotes coverage of diverse teacher-supported motions, while the student-to-teacher direction suppresses unsupported generations. We further introduce differentiable decoded-space kinematic supervision to complement normalized representation matching with constraints in the decoded motion space. At inference, MixiMotion generates a complete motion sequence with a single network evaluation, without teacher queries, iterative sampling, or candidate ranking. On ViMoGen, MixiMotion achieves a semantic alignment score of $0.835$, outperforming the evaluated one-step baselines and approaching the $0.858$ score of its 50-step HY-Motion-1.0-Lite teacher. In blinded human evaluation, MixiMotion obtains an overall rating of $4.33$, compared with $4.50$ for the teacher, while outperforming the evaluated one-/few-step baselines. Meanwhile, generation latency is reduced from $829.58$\,ms to $9.30$\,ms, corresponding to an $89.2\times$ speedup. These results demonstrate an effective quality--efficiency trade-off for strict one-step text-to-motion generation.