Research map / Video / Video generation
VideoWorld 2: Learning Transferable Knowledge from Real-world Videos
Zhongwei Ren, Yunchao Wei, Xiao Yu, Guixun Luo, Yao Zhao, Bingyi Kang, Jiashi Feng, Xiaojie Jin
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 |
|---|---|---|---|
| replace the original LDM decoder with a pretrained video diffusion model (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 0.0 -> 30.3 | Improved |
| stop gradient from the original decoder to latent codes (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 30.3 -> 47.3 | Improved |
| add reconstructed video as ControlNet-like condition (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 47.3 -> 68.8 | Improved |
| increase query embedding length from 1 to 4 (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 41.9 -> 68.8 | Improved |
| increase query embedding length from 4 to 8 (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 68.8 -> 65.0 | Hurt |
| enlarge codebook from 1000 to 64,000 (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 68.8 -> 29.4 | Hurt |
| enlarge codebook from 1000 to 4096 (ablation) | dLDM · CALVIN avg. length | Avg. Len.: 1.87 -> 1.90 | No clear effect |
| use short T=2 context instead of 93 frames (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 68.8 -> 19.1 | Hurt |
| increase context from 93 to 177 frames (ablation) | dLDM · Video-CraftBench paper folding | success rate (%): 68.8 -> 69.0 | No clear effect |
| use VideoWorld 2 latent codes instead of VideoWorld on Craft-only training | VideoWorld · Video-CraftBench paper folding step 7 | success rate (%): 0.0 -> 68.8 | Improved |
| add OpenX pretraining of the latent model | VideoWorld 2 · Video-CraftBench paper folding step 7 | success rate (%): 68.8 -> 72.3 | Improved |
| add OpenX latent pretraining | VideoWorld 2 · CALVIN ABCD->D | Avg. Len.: 2.36 -> 2.88 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- VideoWorld — VideoWorld: Exploring Knowledge Learning from Unlabeled Videos (CVPR 2025) · better on 5 of 5 shared results
- AdaWorld — AdaWorld: Learning Adaptable World Models with Latent Actions (ICML 2025) · better on 1 of 1 shared result
- CoLA — CoLA: Weakly-Supervised Temporal Action Localization With Snippet Contrastive Learning (CVPR 2021) · better on 1 of 1 shared result
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
- VideoWorld: Exploring Knowledge Learning from Unlabeled Videos (CVPR 2025 · similarity 0.93)
- VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation (NeurIPS 2024 · similarity 0.89)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.88)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.88)
- VideoVLA: Video Generators Can Be Generalizable Robot Manipulators (NeurIPS 2025 · similarity 0.88)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 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.
