Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.07500 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911315399802880 |
|---|---|
| author | Liu, Penghui Wang, Jiangshan Shen, Yutong Mo, Shanhui Qi, Chenyang Ma, Yue |
| author_facet | Liu, Penghui Wang, Jiangshan Shen, Yutong Mo, Shanhui Qi, Chenyang Ma, Yue |
| contents | Multi-object video motion transfer poses significant challenges for Diffusion Transformer (DiT) architectures due to inherent motion entanglement and lack of object-level control. We present MultiMotion, a novel unified framework that overcomes these limitations. Our core innovation is Maskaware Attention Motion Flow (AMF), which utilizes SAM2 masks to explicitly disentangle and control motion features for multiple objects within the DiT pipeline. Furthermore, we introduce RectPC, a high-order predictor-corrector solver for efficient and accurate sampling, particularly beneficial for multi-entity generation. To facilitate rigorous evaluation, we construct the first benchmark dataset specifically for DiT-based multi-object motion transfer. MultiMotion demonstrably achieves precise, semantically aligned, and temporally coherent motion transfer for multiple distinct objects, maintaining DiT's high quality and scalability. The code is in the supp. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_07500 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | MultiMotion: Multi Subject Video Motion Transfer via Video Diffusion Transformer Liu, Penghui Wang, Jiangshan Shen, Yutong Mo, Shanhui Qi, Chenyang Ma, Yue Computer Vision and Pattern Recognition Multi-object video motion transfer poses significant challenges for Diffusion Transformer (DiT) architectures due to inherent motion entanglement and lack of object-level control. We present MultiMotion, a novel unified framework that overcomes these limitations. Our core innovation is Maskaware Attention Motion Flow (AMF), which utilizes SAM2 masks to explicitly disentangle and control motion features for multiple objects within the DiT pipeline. Furthermore, we introduce RectPC, a high-order predictor-corrector solver for efficient and accurate sampling, particularly beneficial for multi-entity generation. To facilitate rigorous evaluation, we construct the first benchmark dataset specifically for DiT-based multi-object motion transfer. MultiMotion demonstrably achieves precise, semantically aligned, and temporally coherent motion transfer for multiple distinct objects, maintaining DiT's high quality and scalability. The code is in the supp. |
| title | MultiMotion: Multi Subject Video Motion Transfer via Video Diffusion Transformer |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2512.07500 |