Summary

MotionSpaceFlow, or MSFlow, generates human motion directly from text in full-resolution motion space. A representation-aware flow-matching model jointly processes language and motion without a learned motion encoder or decoder.

Key Features

Text-conditioned human motion.
Direct motion-space generation.
Full-resolution sequences.
No learned motion autoencoder.
Representation-aware noise scaling.
Joint language-motion attention.
Frame-level motion representation.
Released training and inference code.

About MotionSpaceFlow


Animation and motion researchers can work more directly with individual frames and joints rather than only editing a compressed latent sequence. Core capabilities include Text-conditioned human motion, Direct motion-space generation, Full-resolution sequences, and No learned motion autoencoder.


Public code or model weights are linked from the project, with setup instructions for running or adapting the system. Additional characteristics include Representation-aware noise scaling, Joint language-motion attention, Frame-level motion representation, and Released training and inference code.

Get more likes & reach the top of search results by adding this button on your site!

Embed button preview - Light theme
Embed button preview - Dark theme
TurboType Banner

Subscribe to the AI Search Newsletter

Get top updates in AI to your inbox every weekend. It's free!