Compose Image Prompts from a 1,246-Element Library with Skill Prompt Generator

Good image prompts are assembled, not improvised. Skill Prompt Generator is a Claude Code Skills project that composes prompts from a Universal Elements Library of 1,246 elements plus a community corpus of 675 source prompts, routed by 12 specialized domain skills.

Why This Skill Matters

You do not call a Python tool directly — the project is skills-first. Ask for a portrait, poster, or product shot in natural language and Claude Code routes the request to the matching expert skill, which draws on the shared SQLite element library instead of inventing phrasing from scratch. Version 2.0 adds cross-domain queries (database utilization rising from 40.3% to 79.9%, per the README's own statistics), three generation modes — Portrait, Cross-Domain, and Design — and a design system with 37 color schemes plus border and decoration variables, good for 200,000+ combinations. The repo also ships a .codex/ adaptation, so OpenAI Codex CLI users get the same 12 skills. It stands at 1,482 stars on the SkillMap leaderboard.

Installation

Requires the Claude Code CLI and Python 3.8+; Git is optional — the README also documents a Download ZIP install path.

git clone https://github.com/huangserva/skill-prompt-generator.git
cd skill-prompt-generator
pip install -r requirements.txt

After cloning, the 12 skills under .claude/skills/ are recognized by Claude Code automatically.

Real Workflow: A Cinematic Portrait Prompt

1. Install as above, then open Claude Code in the project directory.
2. Type the README's quick-start example — in Chinese, "generate a cinematic Asian woman, Zhang Yimou film style":

生成电影级的亚洲女性，张艺谋电影风格

3. Claude Code identifies the portrait domain and Portrait mode, calls the intelligent-prompt-generator skill, and returns a full prompt. The README's sample output opens with dramatic rim lighting, chiaroscuro, a rich reds-and-golds palette, an 85mm lens, and film grain.

A second workflow exercises v2.0's Design mode. Ask for a warm, cute children's education poster (生成温馨可爱风格的儿童教育海报) and the skill fuses SQLite elements with YAML color variables, returning a complete design spec — color scheme, decorative elements, border style, corner radius — rather than plain prose.

Tips

• Skills are the recommended entry point; the repo also exposes the v2.0 engine (core/cross_domain_generator.py) and the v1.0 IntelligentGenerator API for direct Python use.
• The element library leans portrait-heavy: 502 portrait elements and 208 common photography elements, versus 166 for design, 79 for interior, and 78 for product. Niche domains are thinner — lifestyle carries 4 elements.
• The learning system extracts elements from new prompts, classifies them by domain, and scores reusability, so the library grows as you feed it.
• prompt_framework.yaml defines the portrait framework across seven categories — subject, facial, styling, expression, lighting, scene, technical — with dependency and validation rules.

When Not to Use This

The README describes the system as generating AI image prompts — pair it with your image tool of choice. It needs Python 3.8+ under the hood, so pure-CLI setups without Python will not get far. And outside the supported domains the shelves are thin; a lifestyle brief has only 4 elements to draw on, so expect generic results at the edges.

See the leaderboard for more skills.