Research map / Video / Video generation
STANCE: Controllable Video Generation for Structured Dynamics via Sparse-To-dense ANChored Encoding
ZhiFei Chen, Tianshuo Xu, Leyi Wu, Luozhou Wang, Dongyu Yan, Zihan You, Wenting Luo, Yingcong Chen
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 Dense RoPE motion tokens instead of low-res 2D map (ablation) | STANCE · Small-object subset | Physics-IQ: 31.92 -> 41.83 | Improved |
| use Dense RoPE instead of 2D map (ablation) | STANCE · Regular scenes | Physics-IQ: 43.72 -> 46.89 | Improved |
| add instance control (2D map) to text-conditioned baseline (ablation) | CogVideoX-1.5 · Regular scenes | Physics-IQ: 24.08 -> 43.72 | Improved |
| add joint depth auxiliary stream (ablation) | STANCE · Regular scenes | Physics-IQ: 46.89 -> 49.03 | Improved |
| add joint segmentation instead of depth auxiliary (ablation) | STANCE · Small-object subset | Physics-IQ: 45.63 (depth) vs 45.12 (seg) | Mixed |
| add active-site re-sampling and first-frame RoPE together (ablation) | Dense RoPE · Small + regular merged | Physics-IQ: 31.92 -> 37.83 | Improved |
| add first-frame RoPE to re-sampling (ablation) | Dense RoPE · Small + regular merged | FVD: 58.18 -> 54.63 | Improved |
| use STANCE-5B instead of VLIPP | image-to-video · Held-out 200 clips | Physics-IQ: 36.40 -> 47.62 | Improved |
| STANCE-5B vs 14B Wan-Move | trajectory control · Held-out 200 clips | Physics-IQ: 47.62 vs 47.50 | No clear effect |
| add joint depth stream (ablation) | STANCE · Regular scenes | FVD: 54.63 -> 50.39 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 4 on most shared results.
- MotionPro — MotionPro: A Precise Motion Controller for Image-to-Video Generation (CVPR 2025) · better on 2 of 2 shared results
- Wan-Move-14B — Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025) · better on 1 of 2 shared results
- SG-I2V — SG-I2V: Self-Guided Trajectory Control in Image-to-Video Generation (ICLR 2025) · better on 2 of 2 shared results
- Drag-Anything — Drag Anything: Motion Control for Anything using Entity Representation (ECCV 2024) · 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
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.90)
- LooseControlVideo: Directorial Video Control using Spatial Blocking (ECCV 2026 · similarity 0.89)
- 3D-Aware Implicit Motion Control for View-Adaptive Human Video Generation (CVPR 2026 · similarity 0.89)
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.88)
- Learn the Force We Can: Enabling Sparse Motion Control in Multi-Object Video Generation (AAAI 2024 · similarity 0.88)
- Training-free Camera Control for Video Generation (ICLR 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.
