Comp-I / docs /PHASE2E_STYLE_REFERENCE_GUIDE.md
axrzce's picture
Deploy from GitHub main
338d95d verified
# CompI Phase 2.E: Style Reference/Example Image to AI Art - User Guide
## 🎨 Transform Any Image into AI Art Guidance
Phase 2.E allows you to use **any reference image** (from your device or the web) to guide the style, mood, and composition of your AI-generated art. Upload local files or paste URLs from Google Images, Pinterest, Instagram, or any other source!
## πŸš€ Quick Start
### 1. **Launch the Application**
```bash
streamlit run src/ui/compi_phase2e_streamlit_ui.py
```
### 2. **Basic Workflow**
1. **Enter your text prompt** - Describe what you want to create
2. **Add reference image** - Upload file or paste web URL
3. **Review AI suggestions** - See automatic style analysis
4. **Adjust settings** - Control reference strength and parameters
5. **Generate art** - Create AI art guided by your reference
6. **Download results** - Save images with full metadata
## πŸ“Έ Reference Image Sources
### **Supported Input Methods**
#### πŸ–ΌοΈ **Local File Upload**
- **Formats**: PNG, JPG, JPEG, BMP, TIFF, WebP
- **Size Limit**: Up to 10MB per file
- **Quality**: Higher resolution = better style analysis
#### 🌐 **Web URL Input**
- **Google Images**: Right-click β†’ "Copy image address"
- **Pinterest**: Click image β†’ Copy URL from address bar
- **Instagram**: Use image direct links
- **Art Websites**: DeviantArt, ArtStation, Behance
- **Any Website**: Direct image URLs (.jpg, .png, etc.)
### **URL Examples**
```
βœ… Good URLs:
https://example.com/artwork.jpg
https://pinterest.com/pin/123456789/
https://images.unsplash.com/photo-123/image.jpg
❌ Avoid:
https://website.com/gallery-page (not direct image)
https://social-media.com/post/123 (post page, not image)
```
## πŸŽ›οΈ Interface Guide
### **Main Controls**
#### **Text Input Section**
- **Main Prompt**: Primary description of what you want to generate
- **Style Keywords**: Additional artistic style descriptors
- **Mood/Atmosphere**: Emotional tone and feeling
#### **Reference Image Section**
- **Upload Tab**: Drag & drop or browse for local files
- **URL Tab**: Paste any web image URL
- **Analysis Display**: Real-time style analysis and suggestions
#### **Generation Settings**
- **Reference Strength**: How closely to follow the reference (0.1-0.9)
- **Number of Images**: Generate 1-4 variations
- **Quality Settings**: Inference steps and guidance scale
- **Seed Control**: Random or fixed for reproducibility
### **AI Style Analysis**
When you load a reference image, the AI automatically analyzes:
- **Visual Properties**: Brightness, contrast, color distribution
- **Style Characteristics**: Artistic technique, complexity, mood
- **Suggested Keywords**: Automatically generated style descriptors
- **Enhancement Options**: One-click addition to your prompt
## βš™οΈ Settings Guide
### **Reference Strength Control**
The reference strength determines how closely your generated art follows the reference image:
| Strength | Effect | Best For |
|----------|--------|----------|
| **0.1-0.3** | Loose inspiration, high creativity | Abstract concepts, creative freedom |
| **0.4-0.6** | Balanced style transfer | Most use cases, artistic guidance |
| **0.7-0.9** | Close adherence to reference | Style mimicking, specific looks |
### **Quality vs Speed Settings**
| Setting | Fast (10-15 steps) | Balanced (20-30 steps) | High Quality (40-50 steps) |
|---------|-------------------|------------------------|---------------------------|
| **Time** | 30-60 seconds | 1-2 minutes | 3-5 minutes |
| **Quality** | Good for testing | Recommended | Best results |
| **Use Case** | Quick iterations | Final generation | Professional work |
### **Guidance Scale**
| Scale | Effect | Best For |
|-------|--------|----------|
| **5-10** | More creative, loose interpretation | Artistic freedom, abstract art |
| **10-15** | Balanced prompt following | Most use cases |
| **15-20** | Strict prompt adherence | Precise requirements |
## 🎨 Creative Techniques
### **Effective Reference Selection**
#### βœ… **Good References**
- **Clear artistic style** (paintings, digital art, photography styles)
- **Strong visual identity** (distinctive colors, techniques, moods)
- **Good composition** (well-balanced, not cluttered)
- **High contrast** (clear light/dark areas)
#### ❌ **Challenging References**
- **Cluttered images** with too many elements
- **Low contrast** or very dark/bright images
- **Screenshots** or UI elements
- **Text-heavy** images
### **Prompt Enhancement Tips**
#### **Combine Multiple Styles**
```
Base: "A serene mountain landscape"
+ Style: "oil painting, impressionist"
+ Mood: "golden hour, peaceful"
+ AI Suggestions: "soft brushstrokes, warm colors"
= Enhanced: "A serene mountain landscape, oil painting, impressionist, golden hour, peaceful, soft brushstrokes, warm colors"
```
#### **Layer Your Descriptions**
1. **Subject**: What you want to see
2. **Style**: Artistic technique or medium
3. **Mood**: Emotional atmosphere
4. **Details**: Specific elements or effects
### **Reference Strength Strategies**
#### **Creative Exploration** (Low Strength: 0.2-0.4)
- Use reference for general mood/color inspiration
- Allow AI maximum creative freedom
- Good for abstract or conceptual art
#### **Style Transfer** (Medium Strength: 0.5-0.7)
- Balance between reference and creativity
- Maintain reference style while changing content
- Most versatile approach
#### **Style Mimicking** (High Strength: 0.7-0.9)
- Close adherence to reference technique
- Minimal creative deviation
- Good for specific artistic styles
## πŸ“ Output Management
### **File Naming Convention**
Generated files follow a comprehensive naming pattern:
```
{prompt}_{style}_{mood}_{timestamp}_seed{number}_{REFIMG|NOREFIMG}_v{variation}.png
```
**Example:**
```
magical_forest_fantasy_mystical_20250701_143022_seed12345_REFIMG_v1.png
```
### **Metadata Files**
Each image includes a JSON metadata file with:
- Complete generation parameters
- Reference image information
- AI style analysis results
- Reproducibility data
### **Organization Tips**
- **Create project folders** for different art series
- **Use consistent naming** for easy searching
- **Save metadata** for reproducing successful results
- **Export favorites** to separate collections
## πŸ”§ Troubleshooting
### **Common Issues**
#### **"Failed to load image from URL"**
- βœ… Check URL is a direct image link
- βœ… Try right-clicking image β†’ "Copy image address"
- βœ… Ensure URL ends with .jpg, .png, etc.
- βœ… Test URL in browser first
#### **"Generation taking too long"**
- βœ… Reduce inference steps (try 15-20)
- βœ… Lower image count (try 1-2 images)
- βœ… Check GPU memory availability
- βœ… Restart application if needed
#### **"Poor quality results"**
- βœ… Increase inference steps (try 30-40)
- βœ… Adjust reference strength
- βœ… Improve prompt specificity
- βœ… Try different reference images
#### **"Out of memory errors"**
- βœ… Enable memory optimizations in settings
- βœ… Reduce batch size to 1 image
- βœ… Close other applications
- βœ… Use CPU mode if necessary
### **Performance Optimization**
#### **For Better Speed**
- Use GPU if available
- Enable memory optimizations
- Start with lower inference steps
- Generate fewer images per batch
#### **For Better Quality**
- Use high-resolution reference images
- Increase inference steps (30-50)
- Fine-tune reference strength
- Use descriptive, specific prompts
## 🎯 Best Practices
### **Reference Image Selection**
1. **Choose clear, high-quality images**
2. **Match the style you want to achieve**
3. **Consider color palette and mood**
4. **Test different reference strengths**
### **Prompt Writing**
1. **Be specific about desired elements**
2. **Use artistic terminology**
3. **Describe lighting and atmosphere**
4. **Combine multiple style keywords**
### **Iterative Improvement**
1. **Start with medium reference strength**
2. **Generate multiple variations**
3. **Adjust settings based on results**
4. **Save successful parameter combinations**
### **Workflow Efficiency**
1. **Prepare reference images in advance**
2. **Use consistent naming conventions**
3. **Save metadata for reproducibility**
4. **Organize outputs by project/style**
## πŸš€ Advanced Tips
### **Multi-Style Blending**
- Use reference for base style
- Add contrasting style keywords
- Experiment with different strengths
### **Series Creation**
- Use same reference with different prompts
- Maintain consistent style across images
- Vary only specific elements
### **Style Evolution**
- Start with high reference strength
- Gradually reduce for more creativity
- Create progression series
This guide provides everything you need to master CompI Phase 2.E and create stunning AI art guided by any reference image!