This project provides an experimental AI-powered infinite map generator, creating seamless, neighbor-aware tiles on demand. It targets users needing to generate large, continuous maps efficiently, leveraging Google's Nano Banana model for cost-effective, on-demand content creation.

How It Works

The system breaks down map generation into a grid of tiles, processing them in manageable batches that fit the AI model's input limits. It employs Google's Nano Banana model for generating individual tiles and uses a radial blending technique to seamlessly integrate new tiles with existing ones. A checkerboard background matte is utilized to improve the AI's infill accuracy, encouraging it to fill blank spaces consistently.

Quick Start & Requirements

• Installation: Clone the repository, install dependencies with yarn, copy .env.local.example to .env.local, add your Gemini API key, and start the development server with yarn dev.
• Prerequisites: Node.js 18+, Yarn package manager, and a Google Cloud Platform account with Gemini API access.
• Setup: Obtain a Gemini API key from Google AI Studio.
• Links: Google AI Studio

Highlighted Details

• Generates an infinite(-ish), explorable map with Leaflet-based navigation.
• AI-powered tile generation using Google's Nano Banana model.
• Neighbor-aware generation ensures seamless tile edges.
• Features a local-first architecture with file-based storage.

Maintenance & Community

This project is experimental and serves as a demonstration of the Nano Banana model. The author intends to fix existing bugs but is unlikely to add new features, encouraging users to fork and build upon the project. No specific community channels are listed.

Licensing & Compatibility

The project is released under the MIT License, which permits broad use, including commercial applications and linking with closed-source software.

Limitations & Caveats

This is experimental software and should be used at your own risk. Infill accuracy is estimated at around 75%, often requiring regeneration of tiles for better alignment or if initial renders fail. The codebase was developed rapidly ("vibe coded") and may lack comprehensibility or user-friendliness. Users may incur costs for failed API renders.