𝗩𝗶𝗯𝗲𝗚𝗮𝗺𝗲: 𝗧𝘂𝗿𝗻 𝗡𝗮𝘁𝘂𝗿𝗮𝗹 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝗶𝗻𝘁𝗼 𝗣𝗹𝗮𝘆𝗮𝗯𝗹𝗲 𝗚𝗮𝗺𝗲𝘀

Researchers from Nanjing University, Nanyang Technological University and http://Liblib.ai have introduced VibeGame, a multi-agent framework that builds complete 2D games from natural-language prompts.

How it works:

Uses 8 specialized AI agents, including designers, artists, programmers, auditors and reviewers.
An AI-native game engine allows agents to inspect and control game execution frame by frame.
Multiple review and testing cycles identify and fix gameplay and visual issues.
A self-evolution mechanism saves reusable game templates, modules and development workflows for future projects.

Benchmark results:

Using GPT-6.1-Sol, VibeGame scored 72.99, compared with 68.80 for the Codex baseline, on a 30-task GameCraft-Bench subset.

The interesting part is that VibeGame doesn't just generate game code; it also builds, tests, reviews and improves the playable project.

Research: https://arxiv.org/abs/2610.33106

GitHub: https://github.com/tettethu/VibeGame

Project: https://vibegame.tettet.org

#AI #AIAgents #GameDevelopment #MultiAgentSystems #GenerativeAI