Research map / Video / Video generation
Vid2World: Crafting Video Diffusion Models to Interactive World Models
Siqiao Huang, Jialong Wu, Qixing Zhou, Shangchen Miao, Mingsheng Long
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 action guidance (action dropout plus classifier-free guidance) with Extrapolative weight transfer (ablation) | Vid2World · RT-1 (30k steps) | FVD: 28.6 -> 22.4 | Improved |
| add action guidance with Masked weight transfer (ablation) | Vid2World · RT-1 (30k steps) | FVD: 29.4 -> 25.8 | Improved |
| use Masked instead of Shift weight transfer (ablation) | Vid2World temporal convolutions · RT-1 (30k steps, no guidance) | FVD: 29.9 -> 29.4 | Improved |
| use Extrapolative instead of Shift weight transfer (ablation) | Vid2World temporal convolutions · RT-1 (30k steps, no guidance) | FVD: 29.9 -> 28.6 | Improved |
| use Extrapolative instead of Masked weight transfer (ablation) | Vid2World temporal convolutions · RT-1 (30k steps, with guidance) | FVD: 25.8 -> 22.4 | Improved |
| use Extrapolative instead of Masked weight transfer, without guidance (ablation) | Vid2World temporal convolutions · RT-1 (30k steps, no guidance) | FID: 7.07 -> 7.52 | Hurt |
| fine-tune video diffusion model with causal training and action conditioning versus pre-trained base model | DynamiCrafter · RT-1 | FVD: 237.6 -> 18.5 | Improved |
| Vid2World instead of DIAMOND-HQ | world model · CS:GO | FVD: 368.5 -> 106.6 | Improved |
| Vid2World instead of DIAMOND-HQ | world model · CS:GO | FID: 87.2 -> 17.5 | Improved |
| Vid2World autoregressive versus NWM single-step | world model · RECON | FVD / FID: FVD 59.4 vs 31.2; SSIM 0.481 vs 0.389 | Mixed |
| autoregressive Vid2World versus non-autoregressive Vid2World-NAR (ablation) | Vid2World · RT-1 | FVD: 18.7 -> 18.5 | No clear effect |
| Shift weight transfer may fail on linearly increasing inputs (ablation) | temporal convolution causalization · theory (counter-example) | approximation error | Hurt |
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.
- AVID — AVID: Any-Length Video Inpainting with Diffusion Model (CVPR 2024) · 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
- Learning 3D Persistent Embodied World Models (NeurIPS 2025 · similarity 0.90)
- LongScape: Advancing Long-Horizon Embodied World Models with Context-Aware MoE (NeurIPS 2026 · similarity 0.89)
- VidMan: Exploiting Implicit Dynamics from Video Diffusion Model for Effective Robot Manipulation (NeurIPS 2024 · similarity 0.89)
- AdaWorld: Learning Adaptable World Models with Latent Actions (ICML 2025 · similarity 0.89)
- Diffusion for World Modeling: Visual Details Matter in Atari (NeurIPS 2024 · similarity 0.88)
- WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (NeurIPS 2026 · 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.
