Training Custom Models

Learn how to train custom AI models with your brand style and use them for consistent visual generation

What is a Custom Model?

A custom model allows you to train the AI to generate images in your specific brand style, visual format, or design pattern. By providing example images and descriptions, the AI learns to replicate your unique style for consistent visual generation across all your content.

Step 1: Create a Custom Model

Start by creating a new custom model in your dashboard.

Go to Dashboard → Train Model
Click Create New Model button
Enter Model Name (e.g., 'my-brand-style', 'company-social-posts')
Add Description explaining what this model is for
Select Brand Colors (optional but recommended)
Click Create Model

Note: Your model will be created with 'Draft' status and will be ready to receive training pairs.

Step 2: Add Training Pairs

Training pairs are the heart of your custom model. Each pair consists of a text description and a reference image that shows your desired style.

Recommendation: Add at least 5 training pairs for good results. More pairs (10-15) will improve consistency and quality.

What is a Training Pair?

A training pair teaches the AI by example. It consists of two parts:

Text Description: A clear description of what the image represents (e.g., 'Product launch banner with blue gradient and centered white text')
Reference Image: An actual image showing exactly that style and layout

How to Add Training Pairs

Select your created model from the list
Click Add Training Pair button
Enter Text Description - be specific and detailed
Click Upload Image to select your reference image
Choose Image Type category (infographic, photo, social post, etc.)
Click Add Pair to save
Repeat this process to add 5-10 training pairs

Training Pair Examples

Here are examples of effective training pairs:

Text Description	Image Type	What AI Learns
Product banner with blue gradient background and centered text	Banner/Header	Layout composition, color scheme, text placement
Social media post with brand colors and minimal design	Social Post	Minimalist style, brand color usage, social format
Infographic with icons and data visualization in company style	Infographic	Icon style, data presentation, visual hierarchy
Marketing email header with product photo and call-to-action	Email Template	Email format, CTA placement, product photography style

Step 3: Train the Model

Once you have added sufficient training pairs, you can start the training process.

Ensure you have at least 3 training pairs (5-10 recommended)
Click the Start Training button on your model
Wait for the training process to complete
Monitor the progress bar as it trains
Status will change: Draft → Training → Completed

Note: Training typically takes a few minutes. You can continue using the dashboard while training is in progress.

Step 4: Use Your Custom Model

Once training is complete, your custom model is ready to use in multiple ways.

Method 1: Dashboard Interface

Go to Dashboard → Create Visual
Scroll to Choose Visual Style section
Your custom model appears with a brain icon (🧠)
Click to select your custom model
Enter your prompt describing what you want
Click Generate to create image in your trained style

Method 2: API Integration

Use your custom model programmatically via API

bash

curl -X POST https://yourapp.com/api/v1/generate \
  -H "Authorization: Bearer your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Create a marketing banner",
    "model": "my-brand-style"
  }'

How It Works Behind the Scenes

Understanding the training and generation process helps you create better models.

Training Data Collection

Your training pairs (text + images) are stored and analyzed

AI Analysis

The AI analyzes your examples to identify: visual patterns, color schemes, layout styles, content themes, and design elements

Style Learning

The system learns your unique visual language and brand identity

Generation

When you generate: Your prompt + Learned style = Consistent brand visuals

Visual Style: Colors, gradients, textures, and visual effects
Layout Patterns: Text placement, composition, and spatial arrangement
Design Elements: Icons, shapes, typography style, and graphic elements
Content Format: Infographic, banner, post, presentation, or custom format
Brand Consistency: Your unique visual identity and brand guidelines

Best Practices for Training

Follow these guidelines to create high-quality custom models:

Quality Training Pairs

Use 5-10 high-quality examples minimum
Keep consistent style across all examples
Use high-resolution images (minimum 1024x1024)
Ensure images clearly show your desired style

Descriptive Text

Be specific: 'Blue gradient header with white centered text'
Not vague: 'Nice design' or 'Good image'
Include details about colors, layout, and elements
Describe what makes the image unique to your brand

Consistent Examples

All training images should share similar visual style
Use the same color palette across examples
Maintain consistent layout approach
Keep the same visual language and design principles

Model Naming

Use clear, descriptive names: 'company-social-posts'
Not generic: 'model1' or 'test'
Include purpose: 'client-name-banners'
Make it easy to identify later

Managing Your Models

Learn how to manage, activate, and organize your custom models.

Activate/Deactivate Models

Control which models appear in your workspace

Go to Settings → Custom Models tab
Toggle models on/off using the switch
Inactive models won't appear in playground or API
Useful for testing vs production models

Model Status Types

Draft

Just created, needs training pairs to be added

Training

Currently processing - wait for completion

Completed

Ready to use for image generation

Failed

Training error occurred - contact support if needed

Editing Models

You can add more training pairs to existing models to improve results. Simply select the model and add new pairs, then retrain if needed.

Common Use Cases

Discover how custom models can solve real-world design challenges:

Brand Consistency

Generate all marketing materials in your exact brand style

✓ Maintain visual identity across all content without manual design work

Social Media Management

Create consistent posts across all social platforms

✓ Save time while maintaining professional, on-brand social presence

Product Launches

Maintain visual identity across entire campaign

✓ Quickly generate banners, ads, and promotional materials in consistent style

Team Collaboration

Share brand style with team members and clients

✓ Everyone can generate on-brand visuals without design expertise

Client Work

Create custom styles for different clients

✓ Manage multiple brand identities efficiently in one workspace

Content Marketing

Generate blog headers, infographics, and visual content

✓ Produce high-quality visual content at scale

Tips for Success

Maximize the quality and effectiveness of your custom models:

💡 Start Small, Iterate

Begin with 5 training pairs, test the results, then add more pairs to improve quality

💡 Use High-Resolution Images

Minimum 1024x1024 pixels ensures the AI can learn fine details of your style

💡 Keep Descriptions Detailed

The more specific your text descriptions, the better the AI understands your intent

💡 Test with Simple Prompts First

Start with basic prompts to verify the style is learned correctly before complex requests

💡 Add Variety Within Style

Include different layouts and compositions while maintaining consistent visual style

💡 Document Your Process

Keep notes on what works well for future model training

💡 Regular Updates

Periodically add new training pairs to keep your model current with evolving brand guidelines

Troubleshooting

Common issues and how to resolve them:

Generated images don't match my style

Solutions:

•Add more training pairs (aim for 10-15)
•Ensure all training images are consistent in style
•Make text descriptions more specific and detailed
•Check that training images clearly show your desired style

Training fails or gets stuck

Solutions:

•Ensure you have at least 3 training pairs
•Check that all images uploaded successfully
•Try refreshing the page and starting training again
•Contact support if issue persists

Model not appearing in playground

Solutions:

•Check that training is completed (status: Completed)
•Verify model is active in Settings → Custom Models
•Refresh the page to reload available models
•Ensure you're in the correct workspace

Inconsistent results

Solutions:

•Add more training pairs for better consistency
•Ensure all training examples follow the same style guidelines
•Be more specific in your generation prompts
•Retrain the model after adding new pairs

API Integration

Learn how to use your custom models programmatically.

Note: For complete API documentation, see the API Usage Guide.

POSThttps://yourapp.com/api/v1/generate

Generate images using your custom model

Parameter	Type	Required	Description
prompt	string	Required	Description of what you want to generate
model	string	Required	Your custom model name (exactly as created)

Example Request

bash

curl -X POST https://yourapp.com/api/v1/generate \
  -H "Authorization: Bearer vsk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Create a product announcement banner",
    "model": "my-brand-style"
  }'

Example Response

JSON

{
  "success": true,
  "imageUrl": "https://storage.url/image.jpg",
  "visualId": 789,
  "prompt": "Create a product announcement banner",
  "style": "custom_my-brand-style",
  "isCustomModel": true,
  "customModelName": "my-brand-style",
  "trainingPairsCount": 8
}

Next Steps

Ready to create your first custom model?

Start Training

Go to the Train Model page and create your first custom model

API Documentation

Learn how to integrate custom models into your applications

Generate Visuals

Try generating images with standard models first