Iron Leaf Creative Automation Pipeline

A proof-of-concept creative automation tool that generates consistent social ad creatives across multiple aspect ratios using GenAI-generated backgrounds and product compositing.

Built for the Adobe Forward Deployed AI Engineer take-home exercise.

Features

• Campaign Brief Input: Accept structured campaign briefs with products, target region, audience, and messaging
• Regional Background Generation: Uses DALL-E 3 to generate contextual backgrounds based on target region (10+ regions including US regions and global destinations)
• Product Compositing: Overlays consistent product images onto generated backgrounds (shadows baked into product PNGs)
• Multi-Format Output: Generates creatives in 3 aspect ratios:
  • 1:1 (1080×1080) - Instagram Feed
  • 9:16 (1080×1920) - Stories/Reels
  • 16:9 (1920×1080) - YouTube/Web
• Brand Consistency: Applies campaign message and logo overlay according to brand guidelines
• Streamlit UI: Simple web interface for uploading briefs, previewing results, and downloading outputs

Architecture

Campaign Brief → Parse → For Each Product:
    │
    ├── Generate Regional Background (DALL-E)
    │
    ├── Composite Product onto Background
    │
    ├── Resize to 3 Aspect Ratios
    │
    ├── Apply Brand Overlay (text + logo)
    │
    └── Save to Output Folder

Key Design Decisions

1. Product Compositing vs. Pure GenAI: DALL-E cannot guarantee consistent product appearance across generations. By using product PNGs and compositing them onto generated backgrounds, we ensure the same product appears identically across all regional variations.

2. Background-Only Generation: Prompts are crafted to generate scenic backdrops suitable for product photography. Products are overlaid directly onto the backgrounds.

3. Badge-Style Logo: The logo includes its own background to ensure visibility regardless of the generated background colors/brightness.

4. Separation of Concerns: Brand configuration is separate from campaign briefs, mirroring real-world workflows where brand guidelines are fixed across campaigns.

Project Structure

├── app.py                      # Streamlit web interface
├── requirements.txt            # Python dependencies
├── brand_config.json           # Iron Leaf brand configuration
├── pipeline/
│   ├── models.py               # Pydantic models for validation
│   ├── background_generator.py # DALL-E integration
│   ├── compositor.py           # Product + shadow compositing
│   ├── aspect_ratios.py        # Multi-format resizing
│   ├── brand_overlay.py        # Text + logo overlay
│   └── generator.py            # Main orchestrator
├── assets/                     # Brand images and product assets
│   ├── products/               # Product images (PNGs with baked shadows)
│   │   ├── boots.png
│   │   └── shoes.png
│   ├── cta-button.png          # Call-to-action button
│   └── iron-leaf-logo.png      # Brand logo
├── examples/
│   └── campaign_brief.json     # Example campaign brief
└── output/                     # Generated creatives

Setup

Prerequisites

• Python 3.10+
• OpenAI API key with DALL-E access

Installation

1. Clone the repository:

   git clone <repository-url>
   cd adobe

2. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Create a .env file with your OpenAI API key:

   echo "OPENAI_API_KEY=your_key_here" > .env

Running the Application

streamlit run app.py

The application will open in your browser at http://localhost:8501.

Usage

1. Load Campaign Brief: Use the example brief, upload your own JSON, or create one in the UI
2. Review Products: Verify the product images are loaded correctly
3. Generate: Click "Generate Campaign Creatives" to start the pipeline
4. Preview: Review generated creatives in all aspect ratios
5. Download: Download all creatives as a ZIP file

Example Campaign Brief

{
  "campaign_name": "Trail Season 2026",
  "products": [
    {
      "id": "trailblazer-boot",
      "name": "Trailblazer Hiking Boot",
      "description": "Waterproof leather hiking boot with Vibram sole",
      "product_image": "assets/products/boots.png"
    },
    {
      "id": "summit-jacket",
      "name": "Summit Weatherproof Jacket",
      "description": "3-layer breathable shell with sealed seams",
      "product_image": "assets/products/shoes.png"
    }
  ],
  "target_region": "us_pacific_northwest",
  "target_audience": "Outdoor enthusiasts aged 25-45",
  "campaign_message": "Conquer Every Trail"
}

Available Regions

United States

• us_pacific_northwest - Misty forests with moss and ferns
• us_southwest - Desert landscapes with red rocks and sandstone cliffs
• us_northeast - Autumn forests with fall foliage
• us_rockies - Alpine meadows with dramatic mountain peaks
• us_midwest - Rolling prairie grasslands and woodlands

Global

• alps_europe - European Alps with dramatic alpine peaks and valleys
• scandinavia - Scandinavian wilderness with pine forests and fjords
• patagonia - Windswept plains with jagged mountain peaks
• new_zealand - Lush green mountains and rolling hills
• japan_alps - Forested mountain slopes with misty atmosphere

Limitations & Future Enhancements

Current Limitations

• Compositing Realism: Products are overlaid onto backgrounds and may not perfectly integrate with background lighting
• Font Availability: Uses system fonts; custom fonts require additional setup
• Single Region per Campaign: Current implementation generates one region per run
• Product Shadows: Shadows must be baked into product PNGs; no dynamic shadow generation

Future Enhancements

• Adobe Firefly Integration: Swap to Firefly API when enterprise access is available for better product-scene integration
• Multi-Region Generation: Generate all regional variations in a single run
• Localization: Translate campaign messages for different markets
• Brand Compliance Checks: Validate logo presence, color usage, and text contrast
• A/B Variant Generation: Generate multiple creative variations for testing

Technology Stack

• Python 3.10+: Core language
• Streamlit: Web UI framework
• OpenAI DALL-E 3: Background image generation
• Pillow: Image processing and compositing
• Pydantic: Data validation

License

This project was created as part of a job interview exercise.

---

Iron Leaf is a fictional outdoor gear brand created for demonstration purposes.