Research map / Video / Video generation
CameraCtrl II: Dynamic Scene Exploration via Camera-controlled Video Diffusion Models
Hao He, Ceyuan Yang, Shanchuan Lin, Yinghao Xu, Meng Wei, Liangke Gui, Qi Zhao, Gordon Wetzstein, Lu Jiang, Hongsheng Li
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 |
|---|---|---|---|
| train on dynamic real videos with SfM camera annotations in addition to static data (ablation) | CameraCtrl II · single-clip ablation | Motion strength: 129.40 -> 306.99 | Improved |
| add scale calibration to metric space (ablation) | RealCam dataset · single-clip ablation | TransErr: 0.2121 -> 0.1830; RotErr 2.14 -> 1.74 | Improved |
| add camera trajectory distribution balancing (ablation) | RealCam dataset · single-clip ablation | RotErr: 4.56 -> 1.74 | Improved |
| inject camera only at initial layer instead of every DiT layer (ablation) | CameraCtrl II · single-clip ablation | Motion strength: 247.23 -> 306.99 | Improved |
| use single patchify layer instead of complex CameraCtrl-style encoder (ablation) | CameraCtrl II · single-clip ablation | Geometric consistency: 84.00 -> 86.50 | Improved |
| add joint training with camera-unlabeled video (ablation) | CameraCtrl II · single-clip ablation | Motion strength: 279.82 -> 306.99; TransErr 0.2098 -> 0.1830 | Improved |
| use global first-clip reference frame for relative poses (ablation) | CameraCtrl II video extension · extension ablation | Appearance consistency: 0.8032 -> 0.8654 | Improved |
| keep conditioning clips clean and compute loss only on generated clip (ablation) | CameraCtrl II video extension · extension ablation | FVD: 136.78 -> 112.46 | Improved |
| progressive distillation (NFEs 96 to 16) | CameraCtrl II · distillation | sample time and TransErr: 13.83 s -> 2.61 s; TransErr 0.1892 -> 0.2001 | Mixed |
| one-step APT distillation | CameraCtrl II · distillation | sample time and quality: 13.83 s -> 0.59 s; TransErr 0.1892 -> 0.2500 | Mixed |
| replace MotionCtrl/CameraCtrl with CameraCtrl II | I2V camera-controlled generation · 800-clip eval | FVD: 199.53 -> 73.11 | Improved |
Closest papers by meaning
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.89)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.88)
- Taming Video Diffusion Prior with Scene-Grounding Guidance for 3D Gaussian Splatting from Sparse Inputs (CVPR 2025 · similarity 0.87)
- Reconstruct, Inpaint, Test-Time Finetune: Dynamic Novel-view Synthesis from Monocular Videos (NeurIPS 2025 · similarity 0.87)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.87)
- Modular-Cam: Modular Dynamic Camera-view Video Generation with LLM (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.
