Research map / Video / Video generation
Real-Time Motion-Controllable Autoregressive Video Diffusion
Kesen Zhao, Jiaxin Shi, Beier Zhu, Junbao Zhou, Xiaolong Shen, Yuan Zhou, Qianru Sun, Hanwang Zhang
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 |
|---|---|---|---|
| add RL post-training (GRPO with trajectory reward) (ablation) | AR-Drag · Motion-control benchmark (206 clips) | FVD: 210.35 -> 187.49 | Improved |
| add RL post-training (ablation) | AR-Drag · Motion-control benchmark (206 clips) | motion consistency: 4.12 -> 4.37 | Improved |
| add Self-Rollout training (ablation) | AR-Drag · Motion-control benchmark (206 clips) | FVD: 353.75 -> 187.49 | Improved |
| add Self-Rollout training (ablation) | AR-Drag · Motion-control benchmark (206 clips) | FID: 38.13 -> 28.98 | Improved |
| fine-tune motion control and distill to few-step AR model (ablation) | Wan2.1-1.3B-I2V · Motion-control benchmark (206 clips) | latency (s): 45.72 -> 0.44 | Improved |
| use chunk size 3 instead of frame-wise (chunk 1) (ablation) | AR-Drag · Motion-control benchmark (206 clips) | FID: 28.98 -> 27.47 | No clear effect |
| enlarge KV cache from 7 to 15 or 25 frames (ablation) | AR-Drag · Motion-control benchmark (206 clips) | FID: 28.98 -> 28.96 (15), 28.99 (25) | No clear effect |
| use AR-Drag instead of bidirectional multi-step teacher | AR-Drag · Motion-control benchmark (206 clips) | FVD: 151.46 -> 187.49 | Hurt |
| use Self-Rollout full step-by-step rollout instead of Self-Forcing collapsed KV update | AR-Drag vs Self-Forcing · Motion-control benchmark (206 clips) | FID: 34.47 -> 28.98 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Tora — Tora: Trajectory-oriented Diffusion Transformer for Video Generation (CVPR 2025) · better on 5 of 5 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
- AR-Diffusion: Asynchronous Video Generation with Auto-Regressive Diffusion (CVPR 2025 · similarity 0.90)
- ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation (ICLR 2025 · similarity 0.89)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.89)
- Extrapolating and Decoupling Image-to-Video Generation Models: Motion Modeling is Easier Than You Think (CVPR 2025 · similarity 0.89)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.89)
- Go-with-the-Flow: Motion-Controllable Video Diffusion Models Using Real-Time Warped Noise (CVPR 2025 · 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.
