pic-gen

AI Image Generation & Prompt Optimizer — An OpenClaw Skill that supports multiple image generation models including Qwen Wanxiang (通义万相), Banana (Flux), and DALL-E 3.

Features

• ✨ Prompt Optimization: Convert simple descriptions into professional-grade prompts for any platform
• 🎨 Multi-Model Support: Qwen Wanxiang, Banana/Flux, DALL-E 3 — one Skill to rule them all
• 🔑 API Key Management: Send API keys directly in chat, auto-saved to config
• 💬 Conversational Interface: Step-by-step guidance like talking to a product manager
• 📋 Format Coverage: Midjourney / Stable Diffusion / Flux / DALL-E / Qwen Wanxiang

Supported Platforms

Model	ID	Description
Qwen Wanxiang	qwen	Alibaba Cloud DashScope API, default
Banana / Flux	banana	Open-source Flux models
DALL-E 3	dalle	OpenAI API

Quick Start

1. Install

# Install the skill
clawhub install pic-gen

# Install Python dependencies (required before first use)
pip install -r pic-gen/requirements.txt

2. Configure API Key

Option A (Recommended): Use environment variables — keys never touch the config file:

export DASHSCOPE_API_KEY="sk-xxxxxxxx"
export BANANA_API_KEY="your-banana-key"
export OPENAI_API_KEY="sk-xxxxxxxx"

Option B: Edit pic-gen/config/models.yaml manually:

default: qwen
models:
  qwen:
    enabled: true
    api_key: "your-dashscope-key"
    model: "qwen-image-2.0-pro"

Option C: Send the key directly in chat — the Bot writes it to config automatically.

[!warning] Plaintext config: If you use Options B or C, config/models.yaml will contain your API key in plaintext. Make sure the file is never committed to version control (it is gitignored by default).

export DASHSCOPE_API_KEY="sk-xxxxxxxx"
export BANANA_API_KEY="your-banana-key"
export OPENAI_API_KEY="sk-xxxxxxxx"

3. Start Using

In any OpenClaw-enabled chat (Discord, etc.):

Draw a cat on the moon

Bot will:

1. Optimize the prompt (multi-platform versions shown)
2. Ask which model to use (or use default)
3. Generate and return the image

Project Structure

pic-gen/
├── SKILL.md                  # Skill definition (AI reads this)
├── config/
│   └── models.yaml           # Model config (API keys here)
├── scripts/
│   ├── optimize.py           # Core prompt optimizer
│   ├── generate_qwen.py      # Qwen Wanxiang generator
│   ├── generate_banana.py   # Banana/Flux generator
│   ├── generate_dalle.py    # DALL-E generator
│   └── update_config.py      # Config management tool
└── references/
    ├── midjourney.md         # MJ format guide
    ├── stable-diffusion.md   # SD format guide
    ├── flux.md              # Flux format guide
    └── dalle.md            # DALL-E format guide

Prompt Optimization Strategy

Scene-Aware Lighting

Automatically matches lighting/mood to your scene:

Scene	Qwen Wanxiang	Midjourney
Rainy night	rain beams, wet texture, neon reflections	neon reflections on wet streets
Cyberpunk	neon lights, cyber effects, cold tones	neon + cyberpunk atmosphere
Night / stars	moonlight, starburst, deep night sky	neon + night atmosphere
Sunrise / sunset	golden hour, warm tones, soft backlight	golden hour + warm tones
Forest / nature	natural light, dappled light, fresh vibe	natural sunlight + dappled
Cafe / interior	natural + soft light, cozy	warm interior + cozy

Artist Style Detection

Style	Qwen Wanxiang Boost	Midjourney Boost
Van Gogh / oil painting	bold colors, brushstroke texture	Van Gogh inspired, bold brushstrokes
Ghibli / anime	anime style, soft tones, Ghibli lighting	Studio Ghibli, dreamy atmosphere
Pixar / 3D cartoon	3D animation, Pixar texture	Pixar 3D animation style

Example Output

Input: "Van Gogh-style sunflower field"

🎨 Midjourney:
"Van Gogh-style sunflower field, cinematic photography, shot on Canon EOS R5,
post-impressionist art style, Van Gogh inspired, bold colors,
visible brushstrokes, --ar 16:9 --s 400 --v 6"

🖼️ DALL-E 3:
"Van Gogh 风格的向日葵花田, vivid colors, high detail,
professional photography, highly detailed, perfect composition"

⚡ Qwen Wanxiang:
"梵高风格的向日葵花田，细节丰富，高品质，浓烈色彩，笔触感，后印象派风格，厚涂感"

⚠️ API Key Security

• Never share config files containing real API keys on GitHub, Discord, or any public channel
• config/models.yaml is gitignored — make sure api_key fields are empty before committing
• If a key is leaked, regenerate it immediately in the platform console

Dev Commands

# Optimize prompt (local test)
python3 pic-gen/scripts/optimize.py -i "a cat" -p all

# Generate image
python3 pic-gen/scripts/generate_qwen.py -p "a cat" --download -o ./output

# Config management
python3 pic-gen/scripts/update_config.py show
python3 pic-gen/scripts/update_config.py set-key qwen your-key

Publish to ClawHub

clawhub publish ./pic-gen --slug pic-gen --name "pic-gen" --version 1.0.0

Links

• GitHub: https://github.com/AoturLab/pic-gen
• ClawHub: https://clawhub.com/skills/pic-gen

License

MIT