HKUST and Tencent unveil Möbius: unified motion-language model beats 3B specialists at 0.6B
bdsqlsz · x · 2026-11-07
Researchers from HKUST, Tencent Games Motus, and MIT released Möbius, a unified motion-language model that connects text and motion via shared multimodal attention, supporting motion captioning, text-to-motion generation, motion completion, and instruction-guided editing.
Key results on the HumanML3D test set:
- The 0.6B unified model reaches R@1 of 0.8209 with FID of just 6.13, clearly beating specialized generators (MoMask, MDM, MotionStreamer) and unified baselines like MotionGPT3
- The 1.7B version trained on large-scale data posts R@1 0.7968, outperforming same-scale rivals Go to Zero (3B) and HY-Motion (1B)
Project page, GitHub repo, and arXiv paper are all public.
More from Research
- DSReg: multi-lab preprint disentangles predictive representations without decoders — jiqizhixin · 2026-11-07
- New paper 'Inter-AI Morality' studies how AI systems treat one another — jacyanthis · 2026-11-07
- Steering one GQA key/value group makes a model claim "I am a user" 40 of 40 times — Sauers_ · 2026-11-07
- New research examines the economics of AI for science: winners and bottlenecks — Afinetheorem · 2026-11-07
- Sand3D Explorer lets you browse 106K+ 3D objects with parts, mechanisms and code — victormustar · 2026-11-07
- AI for bio is here: agent usage in Benchling surged through late 2026 — AllThingsApx · 2026-11-07