Research map / Video / Video generation
FlexAM: Flexible Appearance-Motion Decomposition for Versatile Video Generation Control
Mingzhi Sheng, Zekai Gu, Peng Li, Cheng Lin, Hao-Xiang Guo, Yingcong Chen, Yuan Liu
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 |
|---|---|---|---|
| use FlexAM motion video (multi-frequency + depth encoding) instead of DaS 3D tracking video | Motion transfer · Motion transfer | CLIP text-video alignment: 32.14 -> 32.55 | Improved |
| use FlexAM motion video instead of DaS 3D tracking video | Motion transfer · Motion transfer | CLIP temporal consistency: 0.968 -> 0.976 | Improved |
| use FlexAM instead of VACE | Motion transfer · Motion transfer | VBench++ subject/background consistency: 0.86 -> 0.99 | Improved |
| use FlexAM instead of VACE | Motion transfer · Motion transfer | VBench++ aesthetic quality: 0.57 -> 0.64 | Improved |
| use FlexAM instead of Wan2.2 Fun | Motion transfer · Motion transfer | VBench++ imaging quality: 0.68 -> 0.68 | No clear effect |
| use FlexAM instead of VACE | Partial editing · Partial editing user study | preference, lighting consistency (%): 56% vs. 25% | Improved |
| use FlexAM instead of VACE | Partial editing · Partial editing user study | preference, unedited region preservation (%): 51% vs. 28% | Improved |
| use FlexAM 5B single-GPU instead of 14B VACE | Partial editing · Partial editing efficiency | inference time (s): 348s to 91s | Improved |
| use FlexAM instead of task-specific camera methods | I2V camera control · RealEstate10K (100 sequences) | rotation error | Improved |
| use FlexAM instead of Wan2.2 Fun Control Camera | I2V camera control · RealEstate10K (100 sequences) | translation error | Hurt |
| add density-flexible training (stochastic point downsampling) vs DaS (ablation) | FlexAM · Sparse-motion video reconstruction | PSNR, SSIM, FVD | Improved |
| add multi-frequency positional encoding (ablation) | FlexAM motion video · Motion aliasing qualitative | motion identity | Improved |
Closest papers by meaning
- MotionPro: A Precise Motion Controller for Image-to-Video Generation (CVPR 2025 · similarity 0.88)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.88)
- VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models (ICML 2025 · similarity 0.87)
- GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025 · similarity 0.87)
- IM-Zero: Instance-level Motion Controllable Video Generation in a Zero-shot Manner (CVPR 2025 · similarity 0.87)
- Image Conductor: Precision Control for Interactive Video Synthesis (AAAI 2025 · similarity 0.87)
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.
