- Rating
- 1
- Confidence
- 3
- Informativeness
- 1
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
MSFlow is a T2M generator that performs flow-matching directly in motion space, without VAE. It predicts the clean motion (x-prediction), uses a larger Gaussian source scale, and adopts an MMDiT backbone (RA-MMDiT) with a causal attention mask for incremental 263D features and a bidirectional mask for absolute XYZ coordinates. For XYZ, it applies ProjFlow-style projection for zero-shot spatial control. It reports strong results on HumanML3D, SnapMoGen and spatial-control benchmarks.
Critical strengths
Simple and efficient. A compact model (68M) achieves competitive quality at low inference cost, making direct motion-space generation a practical baseline.
Critical weaknesses
-
This paper overclaims its contributions with little novelty. Most components presented as contributions are already established in the community: 1.1 Direct motion-space flow matching without VAE was already done by HY-Motion 1.0, and even earlier by raw-space diffusion models such as MDM. 1.2 Clean-endpoint x-prediction follows JiT, and clean-sample prediction has long been used in motion diffusion (e.g., MDM). 1.3 The core of RA-MMDiT is choosing a causal or bidirectional attention mask according to the motion representation. This is a hyper-parameter choice between two standard masks, both already used in prior work (e.g., causal masking in CMDM, text-to-motion masking in HY-Motion), rather than a new architecture. 1.4 The zero-shot control sampler explicitly adopts ProjFlow's projection.
-
The empirical claims are not convincingly supported. Under the standard HumanML3D evaluator, MSFlow does not even surpass MoMask in FID (0.057 vs. 0.045, App. Table 7).
Questions for authors
- Beyond HY-Motion's direct motion-space MMDiT, what is genuinely new in MSFlow?
- Under the standard HumanML3D evaluator, MSFlow does not surpass MoMask in FID. Why is the MARDM 67D evaluator used as the primary protocol?
4 comments
C. Controlled Fixed-Backbone Comparison Against MDM's -Prediction Objective
To isolate the contribution of our training recipe from that of the architecture, we trained MDM's exact -predicting DDPM objective on our identical 68.8M RA-MMDiT network for 500 epochs:
8/8 method × representation × evaluator rows are complete. FLOW rows report the paper results (Tables 1 & 7); MDM rows report the new 500-epoch evaluations on the identical 68.8M network.
| Representation | Evaluator protocol | Training recipe | Sampler | FID ↓ | Top-3 ↑ |
|---|---|---|---|---|---|
| Incremental 263D | 67D | FLOW (Ours, Paper Table 1) | Heun 50 points | 0.046 ± 0.004 | 0.853 ± 0.003 |
| Incremental 263D | 67D | MDM (New, Same 68.8M Net) | DDPM 1000 steps | 0.291 ± 0.019 | 0.840 ± 0.007 |
| Incremental 263D | standard263 | FLOW (Ours, Paper Table 7) | Heun 50 points | 0.057 ± 0.003 | 0.863 ± 0.003 |
| Incremental 263D | standard263 | MDM (New, Same 68.8M Net) | DDPM 1000 steps | 0.245 ± 0.005 | 0.846 ± 0.001 |
| Global XYZ | 67D | FLOW (Ours, Paper Table 1) | Heun 50 points | 0.038 ± 0.004 | 0.849 ± 0.003 |
| Global XYZ | 67D | MDM (New, Same 68.8M Net) | DDPM 1000 steps | 0.122 ± 0.005 | 0.807 ± 0.005 |
| Global XYZ | standard263 | FLOW (Ours, Paper Table 7) | Heun 50 points | 0.050 ± 0.004 | 0.858 ± 0.002 |
| Global XYZ | standard263 | MDM (New, Same 68.8M Net) | DDPM 1000 steps | 0.167 ± 0.004 | 0.822 ± 0.003 |
On the exact same 68.8M network, replacing the MSFlow formulation with MDM's -DDPM objective degrades FID by 6.3× on 67D ( vs. ) and 4.3× on Standard 263D ( vs. ) for 263D motion, and by 3.2×–3.3× for XYZ motion ( vs. on 67D; vs. on Standard 263D) while requiring 20× more sampling steps (1,000 vs. 50).
Why MSFlow succeeds where MDM and HY-Motion lag:
-
Velocity-Weighted Endpoint Regression (Eq. 5–6): Converting into an ODE velocity field weights endpoint errors by , concentrating supervision on fine kinematic structure near rather than weighting all noise levels equally.
-
Representation-Aware Source Scaling (): Expanding the source noise scale mitigates the extreme spectral anisotropy of uncompressed motion spaces (Proposition 2, Table 13), cutting FID by 58.6% on 263D () and 73.6% on XYZ () compared to unit-scale flow matching (, Table 3).
-
Representation-Aware Multimodal Routing & Cross-Modal Leak Prevention: Viewing RA-MMDiT as a simple toggle between two masks overlooks the structural interaction demonstrated by our 8-model factorial study (see table in Response to
qJT2): applying a causal motion mask inside a standard bidirectional MMDiT (Causal + Fulltext feedback) creates a 2-hop future-to-past causal leak through the text tokens, degrading 263D FID from to (nearly 10× worse). Conversely, applying causal masking to Global XYZ degrades FID from to (41× worse). Co-designing temporal routing, cross-modal feedback blocking, and source-noise scaling for each representation is what enables MSFlow to outperform HY-Motion 1.0 by 9.3× and fixed-backbone MDM by 6.3×.
2. Why the 67D Evaluator Measures Actual Motion Quality, Whereas MoMask Overfits to 196 Discarded Channels in 263D (Weakness 2 & Question 2)
"Under the standard HumanML3D evaluator, MSFlow does not surpass MoMask in FID (0.057 vs. 0.045, App. Table 7). Why is the MARDM 67D evaluator used as the primary protocol?"
- In HumanML3D, Only the 67D Features Construct the Actual 3D Motion, and the Remaining 196 Channels Are Discarded:
As established by MARDM (CVPR 2025), ACMDM (arXiv 2025), and CMDM (CVPR 2026), when the HumanML3D pipeline reconstructs and renders an actual 3D skeletal motion from a 263D feature vector, only the first 67 dimensions (4 root trajectory parameters + root-relative 3D joint positions) are used to recover the 3D human motion. The remaining 196 dimensions (redundant local velocities, IK-approximated 6D rotations, and binary foot-contact labels) are completely discarded during motion recovery and do not directly determine the rendered 3D motion.
- Because the standard 263D evaluator takes 259 channels as input, 192 of its 259 input channels (74%) are redundant features that are thrown away when producing the actual 3D motion.
- In contrast, the 67D evaluator operates exclusively on the exact 67 kinematic features that define the rendered 3D motion, providing a direct measure of true motion quality across both 263D and 66D XYZ representations.
- MoMask's Degradation from 263D FID () to 67D FID () Demonstrates Overfitting to the 196 Discarded Channels: Comparing MoMask and MSFlow across all three HumanML3D evaluation protocols exposes this effect clearly:
| Method | 67D FID (Actual 3D Motion) ↓ | 67D Top-1 / Top-3 ↑ | Std-263D FID ↓ | Std-263D Top-1 / Top-3 ↑ | Std-263D MM-Dist ↓ | 272D FID ↓ | 272D Top-3 ↑ |
|---|---|---|---|---|---|---|---|
| MoMask (Discrete VQ-VAE) | 0.116 (3.1× worse) | 0.490 / 0.786 | 0.045 | 0.521 / 0.807 | 2.958 | 12.232 | 0.846 |
| CMDM (Continuous VAE) | 0.078 | 0.563 / 0.849 | 0.068 | 0.588 / 0.860 | 2.620 | — | — |
| FloodDiffusion (Continuous VAE) | — | — | 0.057 | 0.523 / 0.810 | 2.887 | — | — |
| MSFlow (263D, Ours) | 0.046 | 0.571 / 0.853 | 0.057 | 0.591 / 0.863 | 2.565 | 6.540 | 0.955 |
| MSFlow (XYZ, Ours) | 0.038 | 0.566 / 0.849 | 0.050 | 0.581 / 0.858 | 2.607 | — | — |
- Why does MoMask achieve FID on 263D but degrade to FID on the actual 67D motion? Because MoMask's discrete VQ-VAE codebook is trained to reconstruct all 263 channels, where the 196 unused channels outnumber the 67 actual motion channels by nearly 3-to-1. MoMask overfits to the statistics of those 196 discarded channels, while its actual 3D joint positions suffer from VQ quantization errors, resulting in a real 67D motion FID of , which is 3.1× worse than MSFlow XYZ () and 2.5× worse than MSFlow 263D ().
- Even under the Standard 263D Evaluator (Table 7), MSFlow achieves superior overall performance:
- MSFlow (XYZ) achieves FID, outperforming every continuous-latent, diffusion, and flow-matching model in the literature (MLD V2: , SALAD: , MARDM: , CMDM: , FloodDiffusion: , MDM: , ReMoDiffuse: ), despite generating only 66D joint coordinates without training on the 196 redundant channels.
- On Text-Motion Alignment (Standard 263D), MSFlow substantially outperforms MoMask: MSFlow (263D) surpasses MoMask by +7.0 percentage points in Top-1 R-Precision ( vs. ), +5.6 percentage points in Top-3 ( vs. ), and achieves the #1 best MM-Dist ( vs. MoMask's ) across all 16 methods in Table 7.
- Under the 272D Evaluator (Table 8), where all 272 channels are directly utilized for SMPL animation, MSFlow outperforms MoMask by nearly 2× in FID ( vs. ) and +10.9 percentage points in Top-3 R-Precision ( vs. ).
Summary for Reviewer wSyy: MSFlow outperforms the 1.0B HY-Motion 1.0 model by 9.3× in FID with a 15× smaller open model, outperforms fixed-backbone MDM by 3.2×–6.3× in FID, and outperforms ACMDM's paper raw velocity-prediction model by 3.2× in unconditional FID ( vs. ) as well as our source-scaled clean-prediction ACMDM raw-flow variant by 2.3× in unconditional FID ( vs. ) and by 1.62× in controlled FID ( vs. ) under the identical shared ProjFlow sampler. Meanwhile, MoMask's narrow margin on 263D FID ( vs. ) is an artifact of overfitting to 196 redundant channels that HumanML3D discards during motion recovery. When evaluated on the actual 67D 3D motion, MSFlow outperforms MoMask by 3.1× in FID ( vs. ) while also surpassing MoMask by +7.0% in 263D Top-1 R-Precision. We respectfully ask you to re-evaluate our work based on this evidence and raise your rating.
We thank Reviewer
wSyyfor noting that MSFlow is "simple and efficient" and that "a compact model (68M) achieves competitive quality at low inference cost."Below, we address the two central concerns raised in your review:
1. Methodological & Empirical Differentiation Beyond HY-Motion 1.0 and MDM (Weakness 1 & Question 1)
A. Why Operating in Direct Motion Space Does Not Preclude Novelty
Operating directly in motion space does not, by itself, diminish MSFlow's novelty, just as MLD's introduction of latent motion diffusion does not diminish the contributions of subsequent methods such as MotionLCM, MARDM, SALAD, ACMDM, and CMDM. In both paradigms, the scientific contribution lies in identifying and resolving the fundamental bottlenecks that prevented the paradigm from achieving high fidelity. Prior to MSFlow, direct motion-space generators lagged far behind latent-space models in distributional fidelity; MSFlow is the first direct-space framework to surpass state-of-the-art latent methods.
B. Head-to-Head Comparison: Our Open 68.8M Model Outperforms the Closed 1.0B HY-Motion 1.0 Model by 9.3× in FID (Table 8)
Comparing HY-Motion 1.0 and MSFlow directly highlights the difference:
In Table 8, we evaluated MSFlow directly against HY-Motion 1.0 and its post-trained follow-up UMO on the official 272D HumanML3D benchmark:
Despite being 15× larger (1.0B vs. 68.8M) and trained on proprietary data, HY-Motion 1.0 obtains an FID of 61.035, whereas our 68.8M MSFlow achieves 6.540 (a 9.3× reduction in FID), alongside a +12.5 percentage-point improvement in Top-1 R-Precision (0.667→0.792), outperforming even the 460M fine-tuned UMO-Unified (9.460 FID) across every single metric. These results demonstrate that MSFlow represents a substantial advance in direct motion-space generation, delivering higher motion quality and text alignment with a considerably smaller model.