- Who it helps
- Researchers training interactive video and world models.
- Availability
- Paper first posted October 27 and revised October 31. Dataset download and reuse terms were not verified.
What the dataset contains
The paper reports 69,700 clips and 1,655 hours, including 1,358 hours of continuous two-minute sequences. Camera trajectories and event-aligned captions accompany the clips.
GameVerse supplies longer gameplay sequences for world models ↗
Where it could help
Longer coherent examples could help models learn what persists as a camera moves. The authors report longer generation horizons; the result has not been independently reproduced here.
GameVerse supplies longer gameplay sequences for world models ↗
What you can do
- Read the construction methods and release terms before planning training. Open reference ↗
Where the evidence stops
A dataset paper does not establish a usable game engine or a rights-cleared production asset library.
Sources read. Sources read November 5, 2026 UTC. No third-party code, model inference, account workflow or benchmark executed.
Sources & further reading
Paper · paper
GameVerse supplies longer gameplay sequences for world models ↗
Identifies the subject, documented capability or specific verification limit.
Kang He and Alaya Lab collaborators · page checked 2026-11-05
Coverage note: Independent attention and end-to-end execution remain unverified.
The record over time (2)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Source recheck
GameVerse is a gameplay-video dataset intended to help generated worlds remain consistent over longer sequences.
- · Added to Current
GameVerse is a gameplay-video dataset intended to help generated worlds remain consistent over longer sequences.