Image Generation Factory

This project automates batch image generation using large language models (LLMs) to brainstorm ideas and AI artists to create the images.

Features

• Dynamic Image Generation: Generates images based on meta-prompts using selected AI brains and artists.
• Configurable Workers: Easily switch between different LLM brains (Ollama, Gemini) and image generation artists (Banana, Fooocus).
• Metadata Management: Automatically saves generated image metadata to a CSV file.
• Server Management: Optionally starts and stops Ollama and Fooocus services as needed.

Setup

To get this project up and running, follow these steps:

1. Python Environment

This project requires Python 3.12 or newer. We recommend using uv for environment management.

# Create a virtual environment with Python 3.12
uv venv python=3.12

# Install project dependencies
uv sync

2. Configuration File (.env)

Create a .env file in the root of your project to configure API keys, model paths, and other settings. You can find an example in .env.example. Key settings include:

• ACTIVE_BRAIN: Set to ollama or gemini.
• ACTIVE_ARTIST: Set to banana or fooocus.
• Ollama: OLLAMA__URL, OLLAMA__MODEL
• Fooocus: FOOOCUS__URL, FOOOCUS__CHECKPOINT, FOOOCUS__PATH (the absolute path to your Fooocus installation directory)
• Banana: BANANA__API_KEY, BANANA__MODEL

Example .env:

ACTIVE_BRAIN=gemini
ACTIVE_ARTIST=banana

OLLAMA__URL="http://127.0.0.1:11434"
OLLAMA__MODEL="llama3.2"

FOOOCUS__URL="http://127.0.0.1:8888"
FOOOCUS__PATH="/path/to/your/Fooocus-API"

BANANA__API_KEY="your_banana_api_key"
BANANA__MODEL="gemini-2.5-flash-image"

3. Ollama and Fooocus (if active)

If you set ACTIVE_BRAIN=ollama or ACTIVE_ARTIST=fooocus in your .env file, ensure these services are either running or correctly configured for the project to start them automatically. For Fooocus, make sure FOOOCUS__PATH points to the root of your Fooocus installation where its venv and main.py reside.

Usage

To run the image generation pipeline, execute the main.py script:

uv run main.py

The script will:

1. Brainstorm image ideas using the active_brain (e.g., Ollama).
2. Generate images using the active_artist (e.g., Fooocus) based on the brainstormed ideas and paint configurations from settings.py.
3. Save metadata about the generated images to metadata.csv.
4. Save images to the ./images folder by default.

Customizing Image Generation

• Meta-Prompts: Define your image generation themes in text files within the prompts/meta_prompts/ directory (e.g., niche1.txt). Each file acts as a meta-prompt for the LLM to generate specific image ideas.
• Painting Settings: Adjust parameters like aspect_ratio, image_size, styles, negative_prompt, guidance_scale, N_images, and performance in the settings.py file under the PaintConfig class, or override them via environment variables in your .env (e.g., PAINT__IMAGE_SIZE="1024*1024").