README.md

AutoCaptioner

A tool to automatically

• generate detailed image captions to train higher-quality LoRA and
• optimize your prompts during inference.

What is AutoCaptioner?

AutoCaptioner creates detailed, principled image captions for your LoRA dataset. These captions can be used to:

• Train more expressive LoRAs on Flux or SDXL
• Make inference easy via prompt optimization
• Save time compared to manual captioning or ignoring captioning

Installation

Prerequisites

• Python 3.11 or higher
• Together API account and API key

Setup

1. Create the virtual environment:

   python -m venv venv
   source venv/bin/activate
   python -m pip install -r requirements.txt
   

2. Set your Together API key: TOGETHER_API_KEY

3. Run inference on one set of images:

   python main.py --input examples/ --output output/
   

   Arguments

   • --input (str): Directory containing images to caption.
   • --output (str): Directory to save images and captions (defaults to input directory).
   • --fix_outfit (flag): Indicate if character has one outfit (for consistent descriptions).
   • --batch_images (flag): Process images in batches by category.

Gradio Web Interface

Launch a user-friendly web interface for captioning and prompt optimization:

python demo.py

Features

• High-accuracy image captioning with detailed contextual descriptions
• Consistent character descriptions when using the outfit flag
• Batch processing for large image collections
• Optimized for AI model training datasets
• Web interface for easy use

How It Works

AutoCaptioner leverages the Llama-4-Maverick model through the Together AI platform to:

1. Analyze the visual content of your images
2. Generate detailed, structured captions
3. Save the captions as text files alongside your images

Notes

• Images are processed individually in standard mode
• For large collections, batch processing by category is recommended
• Each caption is saved as a .txt file with the same name as the image

Troubleshooting

• API errors: Ensure your Together API key is set correctly
• Unsupported formats: Only .png, .jpg, .jpeg, and .webp files are supported
• Memory issues: For very large images, try processing in smaller batches

Examples

Original Image	Generated Caption
	"tr1gger photorealistic, mountain landscape, hiking gear and outdoor clothing, standing at viewpoint, awestruck, snow-capped peaks and valley, golden sunset light, wide-angle view"
	"tr1gger photorealistic, urban architecture, business suit with briefcase, walking on sidewalk, focused expression, downtown skyscrapers, evening city lights, street-level view"
	"tr1gger photorealistic, culinary presentation, chef's uniform and hat, plating dish, concentrated, modern restaurant kitchen, soft studio lighting, overhead camera angle"

License

MIT License