LuminaWeaver

Clean, Watermark-Free Descriptions from User Input

LuminaWeaver is a Python package that processes user-submitted text descriptions of images or videos containing watermarks and returns structured, watermark-free descriptions. It uses a language model to reinterpret the content while removing any watermark references, ensuring the output is clean and ready for further applications like content creation or archival.

📦 Installation

Install the package via pip:

pip install luminaweaver

🚀 Usage

Basic Usage (Default LLM7)

from luminaweaver import luminaweaver

user_input = "This image shows a sunset over the mountains with a watermark in the bottom right corner."
response = luminaweaver(user_input)
print(response)

Custom LLM Integration

You can replace the default LLM (ChatLLM7) with any LangChain-compatible model. Below are examples using popular LLMs:

OpenAI

from langchain_openai import ChatOpenAI
from luminaweaver import luminaweaver

llm = ChatOpenAI()
response = luminaweaver(user_input, llm=llm)

Anthropic

from langchain_anthropic import ChatAnthropic
from luminaweaver import luminaweaver

llm = ChatAnthropic()
response = luminaweaver(user_input, llm=llm)

Google Generative AI

from langchain_google_genai import ChatGoogleGenerativeAI
from luminaweaver import luminaweaver

llm = ChatGoogleGenerativeAI()
response = luminaweaver(user_input, llm=llm)

🔧 Parameters

🔑 API Key & Rate Limits

• Default LLM: Uses ChatLLM7 from langchain_llm7 (PyPI).
• Free Tier: Sufficient for most use cases.
• Custom API Key: Pass via api_key parameter or environment variable LLM7_API_KEY.
• Get a Free Key: Register at LLM7 Token.

📝 How It Works

1. Takes a user-provided text description (e.g., "Image of a cat with a watermark").
2. Uses an LLM to reinterpret the description while removing watermark references.
3. Returns a clean, structured output (e.g., "Image of a cat").

📦 Dependencies

• langchain-core
• langchain-llm7 (default)
• re (built-in)

🔧 Development

• GitHub Issues: Report here
• Author: Eugene Evstafev (@chigwell)
• Email: hi@euegne.plus

📜 License

MIT License (see LICENSE).