Sahaj.blogs
13 Nov 2025 · Paper summary

Paper summary: Lumine: An Open Recipe for Building Generalist Agents in 3D Open Worlds

Lumine is an open recipe for a generalist game-playing agent built on a vision-language model: pixels in at 5 Hz, keyboard and mouse out at 30 Hz, completing a five-hour storyline and transferring zero-shot to other games.

Problem

Agents that act in rich 3D worlds for hours, in real time, following natural-language instructions, are far beyond most benchmark agents, which work in short, simplified tasks.

Method

Lumine is presented as the first open recipe for such agents. It uses a human-like interaction loop powered by a vision-language model:

  • perceives raw pixels at 5 Hz
  • outputs precise keyboard and mouse actions at 30 Hz
  • invokes reasoning only when needed, to stay real-time

It’s trained in Genshin Impact.

Key results

From the abstract:

  • completes the entire five-hour Mondstadt main storyline with human-level efficiency
  • follows instructions across exploration, 2D GUI manipulation, collection, combat, puzzle-solving and NPC interaction
  • zero-shot to other games: 100-minute missions in Wuthering Waves and the full five-hour first chapter of Honkai: Star Rail

Limitations

There’s no dedicated limitations section in the extracted text. The demonstrations are in a family of commercial games with similar interaction styles.

#agents#games#vision-language

Sources

  1. Lumine: An Open Recipe for Building Generalist Agents in 3D Open Worlds (arXiv:2511.08892) — arXiv
  2. Hugging Face Daily Papers, 2025-11-13 — Hugging Face
  3. Lumine: An Open Recipe for Building Generalist Agents in 3D Open Worlds (arXiv HTML full text) — arXiv

This is a summary written from the linked public sources. Figures and numbers are as reported there.