Research map / Video / Video generation
MoAlign: Motion-Centric Representation Alignment for Video Diffusion Models
Aritra Bhowmik, Denis Korzhenkov, Cees G. M. Snoek, Amir Habibian, Mohsen Ghafoorian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add motion-specific (flow-supervised) features as alignment target instead of raw VideoMAEv2 features (ablation) | MoAlign · VideoPhy2 | Joint: 23.5 -> 24.9 | Improved |
| add distance-weighted soft TRD loss instead of vanilla TRD (ablation) | MoAlign · VideoPhy2 | Joint: 24.1 -> 24.9 | Improved |
| replace relational TRD alignment with classical REPA loss (ablation) | MoAlign · VideoPhy2 | Joint: 24.9 -> 22.3 | Hurt |
| align at transformer layer 18 versus layer 10 (ablation) | MoAlign · VideoPhy2 | Joint: 23.2 -> 24.9 | Improved |
| align at layer 18 versus layer 20 (ablation) | MoAlign · VideoPhy2 | Joint: 23.8 -> 24.9 | Improved |
| MoAlign alignment instead of plain finetuning on same data | CogVideoX-2B (FT) · VideoPhy2 | Joint: 22.8 -> 24.9 | Improved |
| MoAlign instead of VideoREPA (reimplemented) | CogVideoX-2B · VideoPhy2 | Joint: 23.0 -> 24.9 | Improved |
| MoAlign instead of VideoREPA on semantic adherence | CogVideoX-2B · VideoPhy2 | SA: 26.1 -> 28.8 | Improved |
| finetune with MoAlign on semantic adherence | CogVideoX-2B · VideoPhy (overall) | SA: 49.8 -> 49.3 | Hurt |
| MoAlign instead of VideoREPA on physical commonsense | CogVideoX-2B · VideoPhy (overall) | PC: 37.9 -> 39.4 | Improved |
| MoAlign alignment on VBench Total versus base | CogVideoX-2B · VBench | Total score: 80.6 -> 81.3 | Improved |
| MoAlign alignment on VBench-2.0 Total versus base | CogVideoX-2B · VBench-2.0 | Total score: 54.9 -> 55.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CogVideoX-2B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 6 of 8 shared results
- VideoREPA-2B — VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 2025) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Search (NeurIPS 2025 · similarity 0.90)
- Video Diffusion Models are Training-free Motion Interpreter and Controller (NeurIPS 2024 · similarity 0.90)
- Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models (AAAI 2025 · similarity 0.89)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.89)
- TIV-Diffusion: Towards Object-Centric Movement for Text-driven Image to Video Generation (AAAI 2025 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
