NanmiCoder/open-image-prompts

What it solves

This project provides a local-first, searchable archive of over 36,000 images and 21,000 prompts, designed to help users refine their AI art generation. It solves the problem of finding high-quality, traceable visual references and art directions for image generation models without relying on external cloud-based search tools.

How it works

It uses a local SQLite database to store prompt metadata, translations, and visual labels. Users can interact with this data through a web-based gallery or via two specific "Agent Skills": img-gen-taste, which helps define art direction from a brief, and img-gen-prompts, which retrieves relevant image-prompt pairs. The project separates the application code from the large dataset, allowing users to download the database and image assets separately via GitHub Releases.

Who it’s for

It is for AI artists, prompt engineers, and developers building agentic workflows who need a reliable, offline-capable source of truth for image-prompt associations and aesthetic styles.

Highlights

• Local-first architecture: Runs entirely on your machine with no external API keys or vector databases required.
• Agentic integration: Includes installable "Skills" that allow AI agents to search the archive and provide art direction.
• Transparent dataset: Provides a clear, versioned taxonomy of visual labels and bilingual translations.
• Flexible deployment: Supports local execution via shell scripts, Docker, or integration into existing agent frameworks.