obaydata/world-model-gameplay-recording

---
license: cc-by-nc-4.0
task_categories:
- video-classification
- reinforcement-learning
language:
- en
tags:
- world-model
- gameplay
- action-recording
- gamepad
- keyboard
- video-action-pairs
- game-ai
- video-prediction
- action-conditioned
pretty_name: World Model Gameplay Recording
size_categories:
- n<1K
---

# World Model Gameplay Recording

Action-conditioned gameplay video dataset for world model training. Contains synchronized high-resolution gameplay recordings with frame-accurate input action logs (gamepad, keyboard, mouse) from multiple AAA game titles.

## Dataset Description

- **Homepage:** [obaydata.com](https://obaydata.com)
- **Company:** New Oriental Bay Limited
- **Contact:** [e-mail]

## Games Included

| # | Game | Session ID | Duration | Video Format | Video Size | Input Type |
|---|------|-----------|----------|-------------|------------|------------|
| 1 | **Game Session 1** | fwa0NekU | ~5 min | MKV | 582 MB | Gamepad + Keyboard |
| 2 | **The Legend of Zelda: Tears of the Kingdom** | g4qz1DLq | ~15 min | MP4 (1080p) | 1.29 GB | Gamepad (axis + buttons) |
| 3 | **The Witcher 3: Wild Hunt** | g50N33nG | ~13 min | MP4 (1080p) | 1.07 GB | Keyboard + Mouse |

**Total: ~33 minutes of gameplay, ~2.94 GB of video**

## Dataset Structure

```
fwa0NekU/ # Game Session 1
├── raw_videos/
│ └── 2026-02-11 11-17-33.mkv # Screen recording (~582 MB)
└── raw_meta_data/
└── 2026-02-11_11-17-33/
├── timeline.txt # Session start/stop
├── gamepad_axis_0.txt # Analog stick (5103 events)
├── gamepad_button_0.txt # Button press/release (2168 events)
├── key_0.txt # Keyboard (8 events)
└── mouse_wheel_0.txt # Mouse wheel

g4qz1DLq/ # Zelda: Tears of the Kingdom
├── raw_videos/
│ └── 2026-03-10_23-16-30.mp4 # Gameplay recording (1.29 GB, 1080p)
└── raw_meta_data/
└── 2026-03-10_23-16-30/
├── timeline.txt # Session start/stop
├── gamepad_axis_0.txt # Analog stick (continuous)
├── gamepad_button_0.txt # Button events
├── mouse_move_0.txt # Mouse movement
├── mouse_pressed_0.txt # Mouse clicks
└── mouse_wheel_0.txt # Mouse wheel

g50N33nG/ # The Witcher 3: Wild Hunt
├── raw_videos/
│ └── 2026-03-11_10-31-46.mp4 # Gameplay recording (1.07 GB, 1080p)
└── raw_meta_data/
└── 2026-03-11_10-31-46/
├── timeline.txt # Session start/stop
├── key_0.txt # Keyboard events
├── mouse_pressed_0.txt # Mouse button events
└── mouse_wheel_0.txt # Mouse wheel events
```

## Data Format

### Video
- **Format:** MKV / MP4
- **Resolution:** Up to 1080p
- **Content:** Full-screen gameplay recordings

### Action Logs (plain text, one event per line)

**timeline.txt** — Session boundaries:
```
2026-03-10 23:16:30.389
2026-03-10 23:16:30.474: obs_recording_started
2026-03-10 23:31:32.958
```

**gamepad_axis_0.txt** — Analog stick positions:
```
2026-03-10 23:16:31.709: axis_1,d,-0.105 # Left stick Y-axis
2026-03-10 23:16:31.715: axis_0,d,0.103 # Left stick X-axis
```
Format: `<timestamp>: <axis_id>,<direction>,<value>`

**gamepad_button_0.txt** — Button presses:
```
2026-03-10 23:16:33.996: button_1,d # Button pressed (d=down)
2026-03-10 23:16:34.511: button_1,u # Button released (u=up)
```
Format: `<timestamp>: <button_id>,<d|u>`

**key_0.txt** — Keyboard events:
```
2026-03-11 10:31:50.017: w,KEY_DOWN # W key pressed
2026-03-11 10:31:50.345: w,KEY_UP # W key released
```
Format: `<timestamp>: <key>,<KEY_DOWN|KEY_UP>`

**mouse_pressed_0.txt** — Mouse button events:
```
2026-03-11 10:31:48.948: left,d # Left button pressed
2026-03-11 10:31:49.061: left,u # Left button released
```

All actions are timestamped to millisecond precision for frame-accurate alignment with the video stream.

## Use Cases

- **World Model Pre-training**: Learn environment dynamics from video + action pairs
- **Action-Conditioned Video Prediction**: Predict next frames given current frame + action
- **Game Environment Simulation**: Train neural game engines
- **Game AI / Agent Training**: Offline RL and imitation learning from human gameplay
- **Video Understanding**: Temporal reasoning over complex 3D game environments

## Collection Method

- Gameplay recorded using OBS Studio at up to 1080p
- Input actions logged simultaneously with millisecond-precision timestamps via custom recording software
- All streams temporally synchronized to the same system clock
- Real human gameplay (not scripted or automated)

## Production Data Service

This is a **demo dataset**. We offer large-scale game video collection services:

- **Any game title** — PC, console (via capture card), mobile
- **Hundreds of hours** of synchronized gameplay + action data
- **Custom annotation layers**: game state extraction, object detection, event segmentation
- **Multiple players** for behavioral diversity
- **Monthly capacity:** 100,000+ hours

### Contact

- **Email:** [e-mail]
- **Website:** [obaydata.com](https://obaydata.com)
- **Company:** New Oriental Bay Limited

## Citation

```bibtex
@dataset{obaydata2026worldmodel,
title={World Model Gameplay Recording},
author={OBayData Team},
year={2026},
url={https://huggingface.co/datasets/obaydata/world-model-gameplay-recording},
publisher={Hugging Face}
}
```

## License

[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)