GenImageText

GenImageText

⚠️ IMPORTANT: This skill is NOT an image generator. It adds text to images created by your AI image tools.

This skill solves the common problem of AI-generated images having imperfect or garbled text (especially CJK characters). It separates the workflow into two parts:

1. User generates image using their preferred AI tool (Midjourney, GPT Image 2, Stable Diffusion, Gemini, etc.)
2. This skill adds text to the generated image with perfect typography

Supported AI Image Generators

This skill can work with ANY AI image generation tool, including:

Key Point: Step 2 (Image Generation) uses YOUR AI tool. This skill handles Steps 1, 3, 4, 5.

When to Use This Skill

Use this skill when:

1. User wants to add text to an AI-generated image
2. AI-generated image has garbled text that needs to be fixed
3. Keywords match: poster with text, flowcharts, diagrams, social media graphics with captions
4. User mentions text to be included: "write 'xxx'", "title is", "with 'xxx' text", "saying 'xxx'"

Workflow Overview

Step 1: Prompt Separation        ← This Skill
        ├─ Extract text requirements from user prompt
        ├─ Generate image-only prompt (no text descriptions)
        └─ Output: Image Prompt + Text Requirements

Step 2: Image Generation         ← USER'S AI TOOL ⭐
        ├─ User generates base image using their preferred AI
        │  (Midjourney, GPT Image 2, Stable Diffusion, Gemini, etc.)
        ├─ Use the "image-only prompt" from Step 1
        └─ Output: Clean image without text

Step 3: Image Analysis           ← This Skill
        ├─ Analyze the AI-generated image for safe text placement zones
        ├─ Detect layout structure (for flowcharts)
        └─ Output: Layout suggestions with coordinates

Step 4: Ask User Questions       ← This Skill
        ├─ Ask user 5 questions for customization
        └─ Output: User preferences

Step 5: Text Overlay             ← This Skill
        ├─ Render perfect text on the AI-generated image
        └─ Output: Final image with perfect text

Key Point: Steps 1, 3, 4, 5 use this skill. Step 2 uses user's own AI image generator.

Installation

git clone https://github.com/stephenlzc/GenImageText
cd GenImageText
pip install Pillow numpy

Step 1: Separate Prompt

Use scripts/prompt_separator.py to extract text requirements.

Example

User Input:

"Create a Chinese New Year promotional poster with title 'Spring Festival Sale, Up to 50% Off', red festive atmosphere"

Execution:

from scripts.prompt_separator import separate_prompt

result = separate_prompt(user_input)
# result['image_prompt']: "A festive Chinese New Year promotional poster, vibrant red and gold color scheme, traditional lanterns and plum blossoms decoration, celebration atmosphere, high quality, professional, clean composition, suitable for text overlay"
# result['text_requirements']: {"type": "single_or_few", "text_groups": [{"id": "text_1", "content": "Spring Festival Sale, Up to 50% Off", "semantic_position": "auto"}]}

Output:

• image_prompt: Clean visual description without text
• text_requirements: Structured text data with content and position hints

Step 2: Generate Base Image (User's AI Tool)

⚠️ This step uses USER'S own AI image generator, NOT this skill.

Use the image_prompt from Step 1 to generate the base image using:

• Midjourney - Discord-based AI image generation
• GPT Image 2 (ChatGPT Plus, OpenAI API) - OpenAI's image generator
• Stable Diffusion - Local or web-based generation
• Google Gemini/Imagen - Google's AI image tool
• Adobe Firefly - Adobe's AI image tool
• Microsoft Bing Image Creator - Free, powered by GPT Image
• Flux.1 - Open-source high-quality model
• Other AI image generators - Any tool the user prefers

Important:

• The image should contain NO text (use the cleaned prompt from Step 1)
• Save the generated image to outputs/YYYYMMDD/ folder
• Proceed to Step 3 with the generated image

Step 3: Analyze Image Layout

Use scripts/image_analyzer.py to find optimal text placement.

Example

from scripts.image_analyzer import analyze_image, get_text_placement_suggestions

analysis = analyze_image("generated_image.png", text_requirements)
placements = get_text_placement_suggestions(analysis, text_requirements)

Output includes:

• Safe zones for text (low complexity areas)
• Color scheme analysis
• Suggested text colors for contrast
• Flowchart node positions (if applicable)
• Connection line suggestions

Step 4: Ask User Questions

Before rendering, ask user 5 questions to customize the output:

Question 1: Scene Type

"Please select your image type:"

• Poster/Cover (single/few text, visual focus)
• Flowchart/Step diagram (multiple text, logic sequence)
• Infographic/Data chart (multiple text, information display)
• Diagram/Annotation (multiple text, explanation)
• Other: (please describe)

Question 2: Text Content Confirmation

For single/few text:

"Please confirm text content: '[auto-extracted content]'"

• Content correct
• Needs modification: (please fill in complete content)

For flowchart/multiple text:

"Multiple text detected, please confirm:"

No.	Text Content
1	[Content 1]
2	[Content 2]

• All correct
• Modify No.__: (new content)

Question 3: Font Style

For Simplified Chinese users:

"Please select font style:"

• Modern/Minimalist (Source Han Sans - for: posters, tech, business) ⭐Recommended
• Traditional/Calligraphy (Source Han Serif - for: cultural, formal documents)
• Cute/Cartoon (for: children, fun scenes)
• Artistic/Handwritten (for: personal, creative)
• Other: (custom font path)

For Traditional Chinese users:

"請選擇字體風格:"

• 現代簡約 (Source Han Sans TC - for: Taiwan/Hong Kong, business) ⭐推薦
• 傳統書法 (for: cultural themes, formal occasions)
• 可愛卡通 (for: children, fun scenes)
• 藝術手寫 (for: personal, creative)
• 其他: (custom font path)

For Korean users:

"폰트 스타일을 선택하세요:"

• 모던/현대적 (Noto Sans CJK KR - Korean optimized, business) ⭐추천
• 전통적 (traditional themes, formal documents)
• 귀여운/카툰 (children, fun styles)
• 아티스틱 (personal, creative works)
• 기타: (custom font path)

For English users:

"Please select font style:"

• Modern/Minimalist (Roboto - clean, tech, business) ⭐Recommended
• Traditional/Elegant (Open Sans - formal, readable)
• Playful/Cartoon (fun, children-friendly)
• Artistic/Handwritten (personal, creative)
• Other: (custom font path)

Question 4: Text Position

"Please select text position:"

• Top center (title style)
• Bottom center (movie poster style)
• Center (emphasis)
• Auto-detect best position
• Custom: (e.g., "top-right avoiding face", "left vertical")

For flowcharts:

• Horizontal flow (left→right)
• Vertical flow (top→bottom)
• Branch structure
• Precise control: (describe connection relationships)

Question 5: Effects and Style

"Please select text effects (multiple choice):"

• Add shadow (3D effect)
• Add outline (enhance readability)
• Semi-transparent background box (ensure clarity)
• Add border (flowchart nodes)
• Add connection arrows (flowchart)
• No effects

Additional requirements: (e.g., "gold gradient text", "brand color #FF5733")

Step 5: Render Text Overlay (This Skill)

✅ This step uses this skill to add perfect text to the AI-generated image.

Use scripts/text_renderer.py to overlay text on the image generated in Step 2.

Output File Organization

All output files should be organized in the outputs/ directory with the following structure:

outputs/
└── YYYYMMDD/              # Date-based subfolder (e.g., 20250312/)
    ├── YYYYMMDD_poster_final.png
    ├── YYYYMMDD_flowchart.png
    └── YYYYMMDD_infographic.png

Naming Convention:

• Folder: outputs/YYYYMMDD/ (current date)
• Files: YYYYMMDD_<description>.png
• This makes it easy to manage and find files by date

Example

from scripts.text_renderer import render_text_on_image
from datetime import datetime
import os

# Generate date-based paths
today = datetime.now().strftime("%Y%m%d")
output_dir = f"outputs/{today}"
os.makedirs(output_dir, exist_ok=True)

user_choices = {
    "font_style": "modern",
    "text_size": "auto",
    "effects": ["shadow", "outline"],
    "show_connections": True,
}

output_path = render_text_on_image(
    image_path="generated_image.png",
    output_path=f"{output_dir}/{today}_poster_final.png",
    placements=placements,
    user_choices=user_choices
)

Reference Materials

• Trigger Keywords: See references/trigger_keywords.md for complete list
• Layout Patterns: See references/layout_patterns.md for typography best practices
• Flowchart Symbols: See references/flowchart_symbols.md for diagram standards

Common Scenarios

Scenario 1: Poster with Title

User Request: "Create a movie poster with title 'Interstellar', sci-fi style"

Workflow:

Step 1 (This Skill): 
  ├─ Image prompt: "sci-fi movie poster, space theme..." (no text)
  └─ Text: "Interstellar"

Step 2 (User's AI Tool - Midjourney/GPT Image 2):
  └─ Generate base image using the prompt from Step 1

Step 3-5 (This Skill):
  ├─ Analyze the AI-generated image
  ├─ Detect best placement for title
  └─ Render "Interstellar" with perfect typography

Scenario 2: Flowchart

User Request: "Create a user registration flowchart: 1.Fill info 2.Verify email 3.Complete"

Workflow:

Step 1 (This Skill):
  ├─ Image prompt: "clean workflow diagram, professional blue..."
  └─ Text: ["Fill info", "Verify email", "Complete"]

Step 2 (User's AI Tool):
  └─ Generate diagram image using the prompt

Step 3-5 (This Skill):
  ├─ Detect 3 node positions in the AI-generated image
  ├─ Suggest horizontal flow layout
  └─ Render 3 boxed nodes with connecting arrows and text

Remember: This skill only handles Steps 1, 3, 4, 5. Step 2 is done by user's AI image generator.

Font Handling

The skill uses scripts/text_renderer.py with the following priority:

1. User-provided font path: If specified in Question 3
2. Skill assets: Check assets/fonts/ directory
3. System fonts: Search common system font directories
4. Fallback: Default PIL font

Font Recommendations by Language

Simplified Chinese

Mapping:

• Modern/Minimalist → NotoSansCJKsc-Bold.otf
• Traditional/Calligraphy → NotoSerifCJKsc-Bold.otf

Traditional Chinese

Mapping:

• Modern/Minimalist → NotoSansCJKtc-Bold.otf
• Traditional/Calligraphy → NotoSerifCJKsc-Bold.otf (fallback)

Korean

Mapping:

• Modern → NotoSansCJKkr-Bold.otf

English / Latin

Mapping:

• Modern/Minimalist → Roboto-Bold.ttf
• Traditional/Elegant → OpenSans-Bold.ttf

Output Quality Tips

1. Contrast: Ensure text color contrasts with background
2. Size: Text should be readable at thumbnail size
3. Spacing: Maintain safe margins from image edges
4. Consistency: Use consistent font styles within one image
5. Readability: Use shadow/outline on complex backgrounds

Error Handling

• No text detected: Inform user and proceed with normal image generation
• Font not found: Fall back to system default, notify user
• Text too long: Suggest shorter text or smaller font
• No safe zones found: Suggest adding background box