Design Image Generator

A production-oriented Codex skill for generating posters, product visuals, PPT illustrations, infographics, WeChat cover images, campaign key visuals, and other marketing graphics with Google AI Studio and Volcengine Ark.

Built for repeatable creative production, not one-off prompt experiments.

Highlights

• Generate marketing visuals from reusable prompt systems
• Build prompts from asset, industry, platform, and brand layers
• Reuse presets for common Chinese content and growth scenarios
• Generate directly with Google AI Studio or Volcengine Ark
• Use as a Codex skill or as a standalone Python CLI tool

This project combines:

• A reusable skill definition for Codex
• Prompt templates for Chinese content and marketing scenarios
• Industry and platform preset libraries
• A prompt builder that assembles final prompts from reusable layers
• A one-command image generation workflow for Google AI Studio and Volcengine Ark

What It Does

This skill is designed for practical creative work instead of generic prompt hacking.

It helps you:

• Generate WeChat cover images
• Generate e-commerce product posters and hero visuals
• Generate PPT illustrations and cover images
• Generate infographic base visuals
• Generate campaign KV and banner assets
• Reuse presets for common industries and channels

Core Features

• Template-driven prompt generation
• Chinese creative template library
• Industry-specific prompt layers
• Platform-specific prompt layers
• Brand configuration support
• Preset-based generation for repeated scenarios
• Unified image generation CLI
• Support for Google AI Studio image generation
• Support for Volcengine Ark image generation

Project Structure

design-image-generator/
├── README.md
├── SKILL.md
├── agents/
│   └── openai.yaml
├── assets/
│   ├── brand-config-template.json
│   └── presets/
├── references/
│   ├── chinese-template-library.md
│   ├── industry-template-library.md
│   ├── platform-template-library.md
│   ├── prompt-patterns.md
│   └── providers.md
└── scripts/
    ├── generate_from_template.py
    ├── generate_image.py
    ├── list_presets.py
    └── prompt_builder.py

Installation

Option 1: Use as a Codex skill

Copy the folder into your Codex skills directory:

Copy-Item -LiteralPath .\design-image-generator -Destination "$env:USERPROFILE\.codex\skills\design-image-generator" -Recurse

Then restart Codex so the skill can be discovered.

Option 2: Use as a standalone local tool

You can also run the scripts directly from the project directory with Python 3.

Documentation

• English: README.md
• Chinese: README.zh-CN.md
• Skill instructions: SKILL.md

Requirements

• Python 3.10+
• Access to at least one image provider:
  • Google AI Studio / Gemini image generation
  • Volcengine Ark image generation

Environment Variables

Google AI Studio

Required:

GEMINI_API_KEY=your_google_api_key

Optional:

GOOGLE_IMAGE_MODEL=gemini-3.1-flash-image-preview

Volcengine Ark

Required:

ARK_API_KEY=your_ark_api_key
VOLCENGINE_IMAGE_ENDPOINT_ID=ep-xxxxxxxxxxxxxxxx

Optional:

VOLCENGINE_IMAGE_ENDPOINT=https://ark.cn-beijing.volces.com/api/v3/images/generations
VOLCENGINE_IMAGE_MODEL=ep-xxxxxxxxxxxxxxxx

Note:

• For Volcengine Ark, this project expects a deployed endpoint ID in ep-... form.
• A bare model name is not enough for the current Ark integration path.

Quick Start

1. List available presets

python .\scripts\list_presets.py

2. Build a prompt only

python .\scripts\prompt_builder.py `
  --asset-template wechat-cover `
  --industry saas-b2b `
  --platform wechat `
  --vars-file .\assets\brand-config-template.json `
  --set topic="Prompt Engineering Systems" `
  --set subject="Layered blueprint cards and modular workflow frames" `
  --set core_selling_point="system design, workflow, execution"

3. Preview a full generation request without calling the API

python .\scripts\generate_from_template.py `
  --preset-file .\assets\presets\wechat-cover-methodology.json `
  --vars-file .\assets\brand-config-template.json `
  --set topic="Prompt Engineering Systems" `
  --set subject="Layered blueprint cards and modular workflow frames" `
  --set core_selling_point="system design, workflow, execution" `
  --output-dir .\outputs\wechat-cover `
  --print-prompt `
  --dry-run

4. Generate an image from a preset

python .\scripts\generate_from_template.py `
  --provider google `
  --preset-file .\assets\presets\wechat-cover-methodology.json `
  --vars-file .\assets\brand-config-template.json `
  --set topic="Prompt Engineering Systems" `
  --set subject="Layered blueprint cards and modular workflow frames" `
  --set core_selling_point="system design, workflow, execution" `
  --set color_palette="cream, black, signal red" `
  --aspect-ratio 16:9 `
  --output-dir .\outputs\wechat-cover

5. Generate with Volcengine Ark

python .\scripts\generate_from_template.py `
  --provider volcengine `
  --preset-file .\assets\presets\wechat-cover-methodology.json `
  --vars-file .\assets\brand-config-template.json `
  --set topic="Prompt Engineering Systems" `
  --set subject="Layered blueprint cards and modular workflow frames" `
  --set core_selling_point="system design, workflow, execution" `
  --aspect-ratio 16:9 `
  --output-dir .\outputs\wechat-cover-volcengine

Common Presets

Current bundled presets include:

• wechat-cover-saas
• wechat-cover-methodology
• product-poster-fmcg
• ppt-cover-b2b
• xiaohongshu-beauty
• taobao-3c-hero
• video-account-business
• campaign-kv-tech-event

Use scripts/list_presets.py to inspect them from the command line.

Generated Output

Each generation writes:

• One or more image files
• generation-metadata.json

Typical output:

outputs/
└── wechat-cover/
    ├── image-google-20260421-090000-1.png
    └── generation-metadata.json

Prompt Design Approach

This project follows a systems-style prompt design approach instead of one-off prompt writing.

The typical flow is:

1. Lock the brief
2. Choose the provider
3. Choose the asset type
4. Add industry constraints
5. Add platform constraints
6. Apply brand rules
7. Generate the final prompt
8. Generate the image

Recommended Workflow

For repeated production work:

1. Start from a preset
2. Keep reusable brand defaults in assets/brand-config-template.json
3. Use --dry-run to inspect prompts before paying for generation
4. Save successful parameter combinations as new preset files

Limitations

• Long in-image text is still unreliable for most image models
• Product geometry accuracy is better when reference images are provided
• Volcengine support depends on a valid Ark key and a deployed endpoint ID
• Provider-side rate limits or temporary overload may still cause failures

Roadmap

• Automatic prompt export into output directories
• Result scoring for different asset types
• Batch generation mode
• More presets for verticals and channels

Contributing

Suggested contribution areas:

• New preset files for real production scenarios
• Better provider compatibility and retry logic
• Brand-specific template packs
• Evaluation and scoring workflows for generated assets

License

Add your preferred license before publishing publicly.