etoile-pixtral-description 2025.5.160702

No project description provided

etoile_pixtral_description

etoile_pixtral_description is a Python package designed to extract detailed descriptions from images using advanced language models. It focuses on providing concrete descriptions of visible content characteristics on web pages or other digital media, making it particularly useful for content analysis, accessibility improvements, and SEO enhancements.

Installation

To install etoile_pixtral_description, use pip:

pip install etoile_pixtral_description

Usage

The package provides a function describe_image that generates a detailed description of the content found within an image. It uses the ScreenContentSimplified model for structured output.

Setting Up the Model

from langchain_mistralai import ChatMistralAI

CVISION_MODEL = "pixtral-12b-2409"

llm = ChatMistralAI(
    model=CVISION_MODEL,
    temperature=0,
    max_retries=2,
)

Describing an Image

from etoile_pixtral_description import describe_image

# `screenshot.link` should be a string containing the URL to the image you want to describe.
screenshot_link = "https://example.com/path/to/screenshot.jpg"
result = describe_image(llm, screenshot_link, verbose=True)
print(result)
# Assuming `result` contains a property `description` which holds the detailed description.

This function returns a structured response with a detailed description of each content characteristic on the screen, which is useful for various applications, including digital asset management and accessibility compliance.

Features

• Provides detailed and concrete descriptions of images for enhanced understanding and analysis.
• Utilizes advanced machine learning models to interpret and describe visual content.
• Supports detailed descriptions including the semantics of textual content and the aesthetics of visual elements.

Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page.

License

etoile_pixtral_description is licensed under the MIT License.

Acknowledgements

This package was developed by Evgenii (Eugene) Evstafev as part of the Mistral Étoile suite during the London Hackathon. It aims to provide tools that assist in the detailed analysis of digital content for various professional fields.