Research map / Video / Video generation
SynMotion: Semantic-Visual Adaptation for Motion Customized Video Generation
Shuai Tan, Biao Gong, Yujie Wei, Shiwei Zhang, Zhuoxin Liu, Ke Ma, Yan Wang, Kecheng Zheng, Xing Zhu, Yujun Shen, Hengshuang Zhao
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 |
|---|---|---|---|
| combine semantic dual-embedding learning with visual motion adapters instead of semantic-only (Textual Inversion, DreamBooth) | SynMotion · MotionBench T2V | motion accuracy: 37.56% -> 68.60% (vs DreamBooth) | Improved |
| combine semantic and visual adaptation instead of visual-only (VMC, DMT) | SynMotion · MotionBench T2V | subject accuracy: 38.43% -> 97.67% (vs VMC) | Improved |
| add learnable motion embedding to the baseline HunyuanVideo (ablation) | SynMotion · ablation (qualitative, fox claps) | motion correctness | Improved |
| add learnable subject embedding (ablation) | SynMotion · ablation (qualitative, fox claps) | subject appearance | Improved |
| add embedding refiner R for subject-motion fusion (ablation) | SynMotion · ablation (qualitative, fox claps) | coherence | Improved |
| add motion-aware LoRA adapters A (ablation) | SynMotion · ablation (qualitative, fox claps) | motion magnitude and naturalness | Improved |
| use SynMotion instead of DreamBooth on dynamic degree | SynMotion · MotionBench T2V | dynamic degree: 69.77% -> 88.24% | Improved |
| use SynMotion instead of MotionDirector on motion/subject consistency | SynMotion · MotionBench T2V | motion consistency: 99.01% -> 99.50% | Improved |
Closest papers by meaning
- MotionRAG: Motion Retrieval-Augmented Image-to-Video Generation (NeurIPS 2025 · similarity 0.89)
- VMC: Video Motion Customization using Temporal Attention Adaption for Text-to-Video Diffusion Models (CVPR 2024 · similarity 0.89)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.89)
- VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models (CVPR 2025 · similarity 0.88)
- Motion Prompting: Controlling Video Generation with Motion Trajectories (CVPR 2025 · similarity 0.88)
- Enhancing Motion in Text-to-Video Generation with Decomposed Encoding and Conditioning (NeurIPS 2024 · 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.
