Does MSFlow work in the EU?

Available in the EU. Reportedly released as downloadable code under Apache 2.0 with no geographic restrictions, so it can be run anywhere including the EU; the project page itself gives no explicit availability statement. Last verified 14 Nov 2026.

How much does MSFlow cost?

Open source, open source. Research code reportedly released free under Apache 2.0 (per a creator report; not verified on the project page). Licence: Apache-2.0.

Where was MSFlow shown?

In 1 creator video by AI Search. Most recently by AI Search on 11 Oct 2026, at 2:23 in "New AI enzymes, 700+ math breakthroughs, AI for DNA mutations, room temp semiconductors: AI NEWS".

MSFlow

Website

Text-to-human-motion flow-matching model that generates full-resolution motion directly, with training-free joint constraints.

Available in EU3D & Worlds ResearchModel

Updated 14 Nov 2026EU status checked 14 Nov 2026Last shown 11 Oct 2026

MSFlow at a glance

What is MSFlow?: MSFlow is a model. Text-to-human-motion flow-matching model that generates full-resolution motion directly, with training-free joint constraints.
Does MSFlow work in the EU?: Available in the EU. Reportedly released as downloadable code under Apache 2.0 with no geographic restrictions, so it can be run anywhere including the EU; the project page itself gives no explicit availability statement. Last verified 14 Nov 2026.
How much does MSFlow cost?: Open source, open source. Research code reportedly released free under Apache 2.0 (per a creator report; not verified on the project page). Licence: Apache-2.0.
Where was MSFlow shown?: In 1 creator video by AI Search. Most recently by AI Search on 11 Oct 2026, at 2:23 in “New AI enzymes, 700+ math breakthroughs, AI for DNA mutations, room temp semiconductors: AI NEWS”.

Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.

About

MotionSpaceFlow (MSFlow) is a research model for text-driven human motion generation. Instead of working in a learned, downsampled latent space, it uses flow matching to predict clean motion directly in continuous motion space, with no learned motion encoder or decoder. It introduces representation-aware noise scaling and a Representation-Aware Multimodal Diffusion Transformer (RA-MMDiT) that jointly processes token-level text and full-resolution motion features.

The model supports two motion representations: incremental 263D features (HumanML3D-style) using causal attention, and global XYZ joint coordinates using bidirectional attention. The XYZ variant allows zero-shot, inference-time control of any joint at any frame through projection sampling, without control-conditioned training. The authors report state-of-the-art text-to-motion results across datasets, including HumanML3D and SnapMoGen (native 296D representation), and compare against baselines such as SALAD and ACMDM.

According to a YouTube creator summary, code and training scripts have been released (about 1.1 GB) under an Apache 2.0 license; this was not confirmed on the provided project page, which links no repository. It is aimed at researchers and animation practitioners working with motion synthesis. The vendor/authors are not named in the provided material.

Key features