What it does

This repo collects over 1,100 image-generation prompts, each stored as a Markdown case study with example images. A Node script (scripts/generate-dataset.js) parses the Markdown into a structured JSON file, and a vanilla-JS frontend (index.html) turns that into a browsable, filterable gallery with search and one-click copy. It is essentially a content library with a thin build-and-serve layer on top.

The interesting bit

The prompts are aggressively specific—“80年代末日本动画OVA截图,” “3D轴测微缩模型可视化海报,” “奶凶奶凶的女生”—suggesting the author treats prompt engineering as a craft with its own genres and sub-genres. The auto-tagging and static-site pipeline means the collection stays maintainable even as it scales past a thousand entries.

Key highlights

• 1,145+ prompts and counting, split across paginated Markdown files
• Auto-generated data/prompts.json with source, images, tags, and notes
• Zero-build frontend: any static server serves the gallery after one node command
• Tag-based filtering and keyword search over the full dataset
• Targets multiple models: GPT-4o, GPT-5, Grok, Doubao

Caveats

• The README is mostly a table of contents; prompt quality and model compatibility are not benchmarked or explained
• Social channels (WeChat, X) and a QR code are prominently featured, suggesting community growth is a parallel goal
• No license is visible in the provided sources

Verdict

Worth a bookmark if you generate images with Chinese-language prompts and want a searchable recipe box. Skip it if you are looking for a programmable API, prompt-evaluation framework, or English-first content.