image-prompt-generator

中文

image-prompt-generator 是一个给 coding agent 使用的图片提示词生成 skill 。

它适合处理很短的画面描述，并把抽象想法转换成观众能直接看到的画面细节。输出包含中文维度拆解和可直接用于 AI 画图工具的英文 prompt。

功能

• 将抽象描述转换成具体可见的画面元素
• 自动补全主题、动作、环境、构图、光影、风格、文字等维度
• 避免依赖否定词和空泛概念
• 输出中文拆解和完整英文 prompt
• 可被多种 coding agent 作为 skill、规则文件或自定义指令使用

安装方式

通用安装

1. 下载或复制 SKILL.md。
2. 放到你的 coding agent 支持的 skills、rules、instructions 或 prompts 目录中。
3. 按照对应工具的要求重新加载、重启，或开启一个新会话。
4. 如果你的工具没有固定目录，可以把 SKILL.md 内容作为自定义指令添加。

下载文件：

curl -L https://raw.githubusercontent.com/threeaus/image-prompt-generator/main/SKILL.md -o SKILL.md

克隆仓库：

git clone https://github.com/threeaus/image-prompt-generator.git

Codex 示例

mkdir -p ~/.codex/skills/image-prompt-generator
curl -L https://raw.githubusercontent.com/threeaus/image-prompt-generator/main/SKILL.md -o ~/.codex/skills/image-prompt-generator/SKILL.md

如果已经克隆仓库：

mkdir -p ~/.codex/skills/image-prompt-generator
cp SKILL.md ~/.codex/skills/image-prompt-generator/SKILL.md

安装后，重新打开对应 agent 或开始一个新会话，让它加载这个 skill。

使用方式

在你的 coding agent 中直接提出画图相关需求，例如：

帮我画一张赛博朋克风格的东京街头

或：

生成一张适合做公众号封面的图片 prompt，主题是独立开发者的夜晚工作台

当请求包含“画图”“生成图片”“帮我画”“做张图”“image prompt”“画一张”等表达时，这个 skill 可以指导 agent 生成更完整的图片 prompt。

输出形式

=== 维度拆解 ===
- 主题: ...
- 动作: ...
- 环境: ...
- 构图: ...
- 光影: ...
- 风格: ...
- 文字: ...

=== 完整 Prompt ===
...

English

image-prompt-generator is an image prompt generation skill file for coding agents.

It turns short or abstract image ideas into visible scene details. The output includes a Chinese dimension breakdown and a complete English prompt ready for image generation.

Features

• Converts abstract ideas into concrete visible details
• Fills in subject, action, environment, composition, lighting, style, and text
• Reduces vague concepts and negative phrasing
• Outputs both a Chinese breakdown and a complete English prompt
• Can be used by different coding agents as a skill, rules file, or custom instruction

Installation

Generic installation

1. Download or copy SKILL.md.
2. Put it in the skills, rules, instructions, or prompts location supported by your coding agent.
3. Reload, restart, or start a new session according to your agent's requirements.
4. If your agent has no fixed directory, paste the content of SKILL.md as a custom instruction.

Download the file:

curl -L https://raw.githubusercontent.com/threeaus/image-prompt-generator/main/SKILL.md -o SKILL.md

Clone the repository:

git clone https://github.com/threeaus/image-prompt-generator.git

Codex example

mkdir -p ~/.codex/skills/image-prompt-generator
curl -L https://raw.githubusercontent.com/threeaus/image-prompt-generator/main/SKILL.md -o ~/.codex/skills/image-prompt-generator/SKILL.md

If you already cloned this repository:

mkdir -p ~/.codex/skills/image-prompt-generator
cp SKILL.md ~/.codex/skills/image-prompt-generator/SKILL.md

After installation, restart the relevant agent or start a new session so it can load the skill.

Usage

Ask your coding agent for an image prompt, for example:

Create an image prompt for a cyberpunk street in Tokyo.

Or:

Generate an image prompt for a newsletter cover about an indie developer working late at night.

This skill is intended for requests such as "draw an image", "generate an image", "help me draw", "make an image", "image prompt", and similar image-generation requests.

Output Format

=== 维度拆解 ===
- 主题: ...
- 动作: ...
- 环境: ...
- 构图: ...
- 光影: ...
- 风格: ...
- 文字: ...

=== 完整 Prompt ===
...