DevFest DC AI Challenge - Dynamic Ad Creative Generation Agent

A multi-agent AI system for automated digital ad creative generation using Google's Agent Development Kit (ADK) and Vertex AI.

🎯 Project Overview

This project creates an AI-powered system that automates the creation of digital ad creatives by orchestrating multiple specialized agents. The system takes a product description and target audience as input and generates both ad copy and visual creatives for different platforms.

Examples

ADK UI

Ad

Examples with our own UI - Streamlit

Streamlit UI

Ad

🏗️ Architecture

The project uses a hybrid approach combining:

• Google ADK for agent orchestration and communication
• Vertex AI Imagen for image generation
• Gemini 2.5 Flash for text generation and multimodal processing
• Streamlit for the user interface

Agent Structure

Our multi-agent system consists of:

1. Root Agent - Main orchestrator that coordinates the workflow
2. Copy Generator Agent - Generates persuasive ad copy
3. Image Generator Agent - Creates base images using Vertex AI Imagen
4. Ad Composer Agent - Combines copy and images into complete ad creatives

🚀 Features

• Multi-Platform Support: Generates ads optimized for Facebook, Google, and TikTok
• Automated Workflow: End-to-end generation from brief to final creative
• Visual Intelligence: Platform-aware image generation with appropriate framing and composition
• Interactive UI: Streamlit-based interface for easy interaction
• Downloadable Assets: Generated ads and JSON exports

📁 Project Structure

DevFestDC-AI-Challenge/
├── ad_creative_agent/          # Main agent system
│   ├── agent.py               # Root orchestrator agent
│   └── sub_agents/            # Specialized sub-agents
│       ├── copy_generator/    # Ad copy generation
│       ├── image_generator/   # Image creation
│       └── ad_composer/       # Creative composition
├── streamlit_app.py           # Web interface
├── ads/                       # Final composed ad creatives
├── images/                    # Generated base images
├── image_generation_VertexAI.py  # Vertex AI integration
└── README.md                  # Project documentation

🛠️ Setup & Installation

Prerequisites

• Python 3.8+
• Google Cloud Project with Vertex AI enabled
• Google AI Studio API key

Environment Setup

1. Clone the repository:

git clone https://github.com/your-username/DevFestDC-AI-Challenge.git
cd DevFestDC-AI-Challenge

2. Install dependencies:

pip install google-adk
pip install streamlit

3. Set up environment variables in .env:

GEMINI_API_KEY=your_gemini_api_key
GCP_PROJECT_ID=your_gcp_project_id
GCP_LOCATION=us-central1
GOOGLE_CLOUD_PROJECT=your_gcp_project_id
GOOGLE_CLOUD_LOCATION=global
GOOGLE_GENAI_USE_VERTEXAI=True

4. Authenticate with Google Cloud:

gcloud auth application-default login

🎮 Usage

Streamlit Web Interface

Launch the interactive web application:

streamlit run streamlit_app.py

1. Enter your product description
2. Specify your target audience
3. Select your marketing goal (Awareness, Consideration, Conversion)
4. Click Generate Ads to create 3 variations per platform
5. Download individual images or the complete ads JSON

ADK Web Interface (Local Testing)

Test the agent system using ADK's built-in web interface:

adk web

This provides a local web interface to interact directly with your agents for development and testing.

Agent System Integration

Use the ADK agents directly in your code:

from ad_creative_agent.agent import root_agent

# Initialize the root agent
response = root_agent.send_message(
    "Generate ad copy and images for a fitness tracker targeting young professionals"
)

🔧 Core Technologies

• Google Agent Development Kit (ADK) - Agent framework
• Vertex AI Imagen - Image generation
• Gemini 2.5 Flash - Language model for copy generation
• Streamlit - Web interface
• Pillow (PIL) - Image processing

📋 Implementation Status

✅ Completed Features

• Multi-agent system using Google ADK
• Ad copy generation
• Vertex AI Imagen integration for image generation
• Platform-aware visual prompt engineering
• Streamlit web interface
• Image composition and text overlay
• Downloadable assets

🚧 Planned Features (From Original Blueprint)

• LangGraph workflow orchestration
• Advanced evaluator agent with quality scoring
• Analytics agent with performance insights
• Feedback loop for iterative improvement
• A/B testing capabilities

🤝 Contributing

This project was developed for the DevFest DC AI Challenge. Feel free to fork and extend the functionality!

🙏 Acknowledgments

• DevFest DC organizers for the AI Challenge
• Google Cloud for Vertex AI and ADK platform
• The open-source community for supporting libraries