openclaw-image-generator（OpenClaw 图像生成器）

OpenClaw Image Generator

Generate production-ready images using this pipeline:

1. Tiny Stable Diffusion generates background image
2. Pillow renders optional title/subtitle text with customizable styling
3. Output images can be used in any system (blogs, social media, presentations, documentation, etc.)

Trigger Conditions

Use this skill when user intent includes:

Image Generation:

• "Generate/create covers, posters, illustrations"
• "Create cover images for articles, blogs, videos, or social media"
• "Generate background images for presentations or slides"
• "Generate pure background images (without text)"
• "Create branded or product promotional images"

Text & Design Customization:

• "Add title or text to images"
• "Adjust text color, size, and positioning"
• "Create multiple text style variations"
• "Improve text and background integration"

Batch & Optimization:

• "Batch regenerate visual assets"
• "Generate multiple images with different styles at once"
• "Improve AI image generation prompt quality"
• "Generate cover images for multiple topics"

Technical:

• "Tiny SD / Stable Diffusion + Pillow"
• "Generate high-quality images with AI"
• "Need fast image generation"

Repository Assumptions

• Script path: scripts/generate_image.py
• Batch helper: scripts/batch_generate.py
• Jobs config: scripts/jobs.showcase.json (for batch mode)
• Output directory: customizable via --output parameter

If these paths differ, adapt commands to actual repository paths.

Dependency Setup

Install dependencies from bundled requirements:

pip install -r scripts/requirements.txt

Platform Support:

• macOS with Apple Silicon: uses MPS (Metal Performance Shaders)
• NVIDIA GPUs: uses CUDA
• CPU fallback: supported but slower

Backend selection is automatic at runtime.

Standard Workflow

1) Understand the visual requirements

For each image request, gather:

• Target subject or theme
• Desired mood, style, or atmosphere
• Key visual elements
• Text content (title, subtitle) if any

Then create a one-line summary:

• What: Main subject (e.g., "wildlife photography", "urban street", "modern architecture")
• Action: What's happening (e.g., "capturing a moment", "conveying motion", "showing details")
• Visual style: How to depict it (e.g., "cinematic", "photorealistic", "minimalist")
• Mood: Tone and atmosphere (e.g., "dramatic lighting", "warm and cozy", "dynamic energy")

2) Create an English language prompt

Always write prompts in English (Stable Diffusion models only understand English).

Recommended structure:

[subject], [setting/environment], [lighting/mood], [style/quality], [composition details], no text, 4k

Example: majestic deer in misty forest, golden morning light, photorealistic wildlife photography, bokeh background, 4k

Prompt best practices for Tiny SD:

✅ DO:

• Use concrete, specific nouns ("red sports car", "turquoise ocean waves", "luxury modern house")
• Include lighting descriptions ("golden hour", "dramatic sunset", "cinematic lighting")
• Specify mood/atmosphere ("cozy", "dynamic", "serene", "energetic")
• Keep 5-8 key descriptive elements
• Always include no text, 4k

❌ AVOID:

• Chinese or non-English text in prompts
• Too many unrelated objects
• Conflicting visual styles
• Asking the model to render readable text
• Overly complex or abstract descriptions

Negative prompt is handled by the script; do not duplicate unless asked.

3) Generate the image

Text overlay is optional:

• Provide --title (and optional --subtitle) for text-overlaid images
• Omit both to generate background image only

Required parameter: --prompt (no built-in style templates)

Single image with text overlaid:

python3 scripts/generate_image.py \
  --prompt "<English prompt>" \
  --title "<Title text>" \
  --subtitle "<Optional subtitle>" \
  --output "outputs/image.png" \
  --steps 28 \
  --seed 42 \
  --position bottom

Background image only:

python3 scripts/generate_image.py \
  --prompt "tropical beach with turquoise water, white sand, palm trees, 4k" \
  --output "outputs/beach.png"

With custom text styling:

python3 scripts/generate_image.py \
  --prompt "modern luxury house, sunset lighting, contemporary architecture, 4k" \
  --title "Dream Home" \
  --title-color "255,200,50,255" \
  --title-size 60 \
  --outline-width 3 \
  --outline-color "0,0,0,200" \
  --output "outputs/house.png"

Parameter tuning guide:

4) Verify output quality

Inspect each generated image for:

• ✓ Text readability (good contrast with background)
• ✓ Text positioning (no overflow, proper alignment)
• ✓ Visual coherence (style consistency across batch)
• ✓ No artifacts or degradation
• ✓ Output file size and format

Batch Generation Mode

For generating multiple images at once:

Step 1: Create a JSON configuration file (e.g., jobs.json):

{
  "jobs": [
    {
      "name": "wildlife-photo",
      "prompt": "majestic deer in misty forest, golden morning light, photorealistic, 4k",
      "title": "Wild Beauty",
      "subtitle": "Nature's Elegance",
      "output": "outputs/01-wildlife.png",
      "seed": 101,
      "steps": 28,
      "position": "bottom"
    },
    {
      "name": "car-photo",
      "prompt": "red sports car on mountain road, sunset lighting, motion blur, 4k",
      "title": "Speed & Power",
      "output": "outputs/02-car.png",
      "seed": 202,
      "steps": 28,
      "position": "top"
    }
  ]
}

Step 2: Run batch generation:

python3 scripts/batch_generate.py \
  --jobs jobs.json

Step 3: Verify all outputs were generated successfully and report paths.

Prompt Quality Rules

✅ Effective prompt elements:

• Concrete subjects: majestic deer, luxury yacht, modern skyscraper, tropical beach
• Specific lighting: golden hour, dramatic sunset, volumetric light, cinematic glow
• Color palettes: deep blue and cyan, warm golden tones, emerald and teal
• Composition: wide angle, shallow depth of field, symmetrical framing, bokeh background
• Quality markers: photorealistic, 4k, masterpiece, professional photography

❌ Things to avoid:

• Non-English text in prompts
• Overly many unrelated objects
• Contradictory visual styles (e.g., "minimalist cyberpunk")
• Requesting readable text/numbers in images
• Generic descriptions without specifics

Troubleshooting

Issue: Generated image doesn't match prompt

• ✓ Shorten the prompt and move main subject to the beginning
• ✓ Increase --steps to 30-35 for more detail
• ✓ Try different seeds: 42, 101, 202, etc.
• ✓ Check that all words are spelled correctly and in English

Issue: Text overlay looks misaligned or illegible

• ✓ Keep --position bottom unless you need centered or top text
• ✓ Reduce --title-size if text is truncated
• ✓ Try --outline-width 0 if outline is too thick
• ✓ Use a simpler background prompt (fewer objections)
• ✓ Ensure sufficient contrast between text color and background

Issue: Colors or style incorrect

• ✓ Verify --title-color and --outline-color RGBA values
• ✓ Re-check English prompt spelling and grammar
• ✓ Regenerate with different seed values

Issue: File path not found

• ✓ Verify --output path exists or use auto-creation (os.makedirs)
• ✓ Use absolute path or project-relative path consistently
• ✓ Check file permissions in output directory

Response Style When Using This Skill

When generating images, provide clear feedback:

1. Item summary (1 line describing the image concept)
2. Final prompt used for generation
3. Output path where image was saved
4. Configuration notes (seed, steps, text styling applied)
5. Status (success/failure with any issues)

Example output:

✅ Wildlife Photography - Majestic deer in forest
Prompt: "majestic deer standing in misty forest, golden morning light..."
Output: outputs/01-wildlife.png
Config: seed=101, steps=28, title="Wild Beauty", position=bottom