magic-photobooth

Magic Photobooth

A personalized AI image generation product that can create your avatars from a photo or selfie.

✨ Transform Your Photos into Magical Avatars

Magic Photobooth harnesses cutting-edge AI technology to transform photos into extraordinary digital avatars. By providing a dataset of photos, you can train the AI to generate stunning images in any style, setting, or scenario imaginable.

🌟 Key Features

• Personalized AI Models: Custom-trained on your dataset for truly personalized results
• Unlimited Creative Possibilities: Generate images as superheroes, historical figures, or in fantasy worlds
• Animated Avatars: Create short animated videos of your digital twin
• Enterprise-Grade Technology: Powered by state-of-the-art Flux AI models and ZenML's MLOps framework
• Fast Processing: Optimized for quick generation without sacrificing quality

🎭 How It Works

Magic Photobooth uses an advanced AI technique called DreamBooth to create a personalized version of powerful image generation models. Here's how the magic happens:

1. Dataset Preparation: Prepare a collection of 5-10 clear photos and host them on a compatible storage location (your artifact store).
2. Configuration: Update the configuration file with your dataset location and subject details
3. Model Training: The system fine-tunes a custom model specifically for your dataset
4. Avatar Generation: The system generates images based on customizable text prompts

Behind the scenes, Magic Photobooth employs Low-Rank Adaptation (LoRA) technology to efficiently customize the Flux image generation model. For inference, we use the optimized Flux-Schnell model to deliver high-quality results at impressive speeds.

🚀 Getting Started

Prerequisites

• Python 3.8+
• GPU access (for model training)
• ZenML installed and configured
• Hugging Face account (for hosting your dataset)

Quick Start

1. Install Magic Photobooth:

   # Clone the repository
   git clone https://github.com/zenml-io/zenml-projects.git
   
   # Navigate to Magic Photobooth
   cd zenml-projects/flux-dreambooth
   
   # Install dependencies
   pip install -r requirements.txt

2. Prepare your dataset:

   • Create a collection of 5-10 clear photos
   • Upload them to your artifact store
   • Update the instance_example_dir in configs/k8s_run_refactored_multi_video.yaml to point to your dataset

3. Customize your subject:

   • Update the instance_name and class_name in the configuration file to match your subject
   • For example: instance_name: "john_doe" and class_name: "man"

4. Customize generation prompts:

   • Edit the prompts in the batch_inference step in k8s_run.py to specify the styles and scenarios you want
   • For example: "A photo of {instance_phrase} as a superhero" or "A photo of {instance_phrase} in ancient Rome"

5. Explore the interactive tutorial:

   # Open the guided walkthrough notebook
   jupyter notebook walkthrough.ipynb

☁️ Cloud Deployment

Magic Photobooth requires GPU resources for optimal performance. We recommend deploying on a cloud infrastructure:

1. Set up your cloud environment using our 1-click deployment guide for AWS, GCP, or Azure.

2. Configure your GPU quotas to ensure sufficient resources for model training and inference.

3. Run the pipeline using your preferred orchestrator:

   # For Kubernetes environments
   python k8s_run.py
   
   # For Modal
   python modal_run.py

📊 Results & Showcase

Magic Photobooth generates two types of personalized content:

1. Image Galleries: Composite images showcasing your avatar in various styles and scenarios
2. Animated Clips: 3-second videos bringing your static images to life using Stable Video Diffusion technology

🛠️ Technical Details

Magic Photobooth is built on a robust MLOps architecture:

├── configs/               # Configuration profiles for different environments
├── assets/                # Sample outputs and demonstration media
├── k8s_run.py             # Kubernetes deployment script with customizable prompts
├── modal_run.py           # Modal cloud deployment script
└── walkthrough.ipynb      # Interactive tutorial notebook

🔮 Use Cases

• Social Media Content: Create eye-catching profile pictures and posts
• Digital Marketing: Generate custom branded imagery featuring specific individuals
• Creative Projects: Visualize subjects in fictional scenarios or historical periods
• Personal Avatars: Create unique avatars for gaming or virtual worlds

📚 Documentation

For learning more about how to use ZenML to build your own MLOps pipelines, refer to our comprehensive ZenML documentation.