Gemini Image Generation Batch

A web application that uses Google's Gemini 2.0 Flash experimental image generation model to create images based on text prompts and reference images.

Features

• Generate multiple images with a single text prompt
• Optionally include reference images to guide the style and content
• Run multiple generations in parallel for variations
• All generated images are automatically saved with timestamped filenames

Project Structure

The project is organized as follows:

├── app.py                 # Main application entry point
├── utils/                 # Utility modules
│   ├── __init__.py        # Package initialization
│   ├── async_utils.py     # Async operations utilities
│   ├── file_utils.py      # File and directory utilities
│   ├── gemini_utils.py    # Gemini API interaction utilities
│   ├── logging_utils.py   # Logging utilities
│   └── ui_utils.py        # Gradio UI utilities
├── outputs/               # Generated images output directory
├── requirements.txt       # Project dependencies
└── .env                   # Environment variables (API keys)

Setup

1. Create a .env file with your Gemini API key:

GEMINI_API_KEY=your_api_key_here

2. Install the required packages:

pip install -r requirements.txt

3. Run the application:

python app.py

For custom number of reference image inputs:

python app.py --num-images 6

To create a shareable link:

python app.py --share

Usage

1. Enter a detailed text prompt describing the image you want to generate
2. (Optional) Upload one or more reference images to guide the generation
3. Set the number of parallel generations (1-8)
4. Click "Generate Images"
5. All images will be saved in the "outputs" directory

Using the App

1. Enter your prompt text describing the image you want to generate
2. Optionally upload up to 4 reference images to guide the style or content
3. Adjust the slider for how many parallel generations you want (1-8)
4. Click "Generate Images" to start the process
5. The generated images will appear below the input form in a grid layout

The app uses Gemini 2.0 Flash experimental image generation model and can generate up to 8 variations simultaneously.

Batch Processing

The app supports batch processing through the "Number of parallel generations" slider:

• Select a value between 1 and 8 to generate that many versions of your prompt simultaneously
• All images are generated in parallel using the same prompt and reference images
• Results are displayed in a grid layout with up to 4 images per row
• Text responses (if any) will be collected in the "Generation Details" section

This feature is useful for:

• Exploring different variations of the same prompt
• Finding the best result among several options
• Saving time compared to generating images one by one

Advanced Usage

Command-line Arguments

The app supports several command-line arguments:

• --num-images N: Change the number of reference image inputs (default: 4)
• --share: Create a shareable public link (useful for demos)

Example:

# Run with 2 reference image inputs
python app.py --num-images 2

# Run with 6 reference image inputs
python app.py --num-images 6

# Create a public shareable link
python app.py --share

Examples

The app includes several example prompts to get you started. Click on any example to load it into the prompt field.

Limitations

• The Gemini 2.0 model has a maximum context size, so very large images might cause errors
• Generation can take a while depending on server load and how many parallel runs you request
• Running 8 generations simultaneously may be resource-intensive depending on your system
• The model may occasionally return text responses instead of images for certain prompts