Research map / Video / Video generation
GameVerse: A Minute-Scale Gameplay Dataset for Long-Horizon Interactive World Modeling
Kang He, Wenshuo Peng, Chuanhao Li, Zihui Gao, Xiaojie Xu, Zhengyuan Lin, Yuzhe Ding, Donghong Ji, Kaipeng Zhang, Yongtao Ge
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces GameVerse, a minute-scale game video dataset designed to support long-horizon, camera-controllable, and semantically structured interactive world modeling.
Problem
Existing video datasets do not simultaneously support long-horizon dynamics, precise camera control, and structured semantic representations needed for interactive world models.
Approach
A large-scale game video dataset with minute-scale clips that pairs long-horizon gameplay with camera control and structured semantic annotations for training interactive world models.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune HY-World 1.5 on about 600 hours of GameVerse | HY-World 1.5 · VBench, 60s generation | Subject Consistency: 66.92 -> 68.45 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · VBench, 60s generation | Overall Consistency: 14.48 -> 15.69 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · WorldScore, 60s generation | WS-Static: 53.33 -> 55.13 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · WorldScore, 60s generation | 3D Consistency: 66.94 -> 73.42 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · WorldScore, 60s generation | Camera Control: 28.58 -> 28.90 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · reference-frame similarity | CLIP similarity at 30s: 0.609 -> 0.642 | Improved |
| fine-tune HY-World 1.5 on GameVerse | HY-World 1.5 · Subject Consistency by activity | Subject Consistency: improvements of 1.15, 1.85, 2.49, and 1.58 points | Improved |
| fine-tune LTX-Video 2.3 on GameVerse with camera and caption conditioning (versus LingBot, different architecture) | LTX-Video 2.3 · WorldScore, 60s generation | Camera Control: 24.33 vs LingBot 31.69 | Hurt |
| GameVerse fine-tuned HY-World versus LingBot on perceptual quality | HY-World-FT · VBench, 60s generation | Imaging Quality: 49.34 vs LingBot 56.70 | Hurt |
| use Wan2.2 instead of LTX-Video 2.3 backbone (no fine-tuning) (ablation) | original backbones · VBench, 5/10/15s | Aesthetic and Imaging Quality: average advantages of 1.87 and 11.64 points | Improved |
Closest papers by meaning
- DynamicVerse: A Physically-Aware Multimodal Framework for 4D World Modeling (NeurIPS 2025 · similarity 0.89)
- GameVerse: Can Vision-Language Models Learn from Video-based Reflection? (ICML 2026 · similarity 0.87)
- GameGen-X: Interactive Open-world Game Video Generation (ICLR 2025 · similarity 0.87)
- From Zero to Hero: Training-Free Custom Concept Spawning in World Models (NeurIPS 2026 · similarity 0.87)
- The Matrix: Infinite-Horizon World Generation with Real-Time Moving Control (NeurIPS 2025 · similarity 0.87)
- Trimming the Long-Tail of Visual World Modeling Evaluation (NeurIPS 2026 · similarity 0.86)
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.
