text-humanizer README

The project's README from github.com/kurelas/text-humanizer. As one markdown file: README.md.

text-humanizer

image

License: MIT Python

English | 中文

Overview

text-Humanizer is a completely free open-sourced project designed to humanize AI-generated text through a multilingual LLM-powered rewriting pipeline. By leveraging large language models for semantic-preserving text transformation, the system rewrites content with more natural sentence structures, varied linguistic patterns, and improved stylistic diversity while maintaining the original meaning across multiple languages.

Features

How it works

Step 1 - LLM Rewrite (DeepSeek)

Text is rewritten using the DeepSeek LLM to generate a semantically equivalent version with modified sentence structures, phrasing, and information flow. During this stage, the model also translates the content into Chinese as an intermediate representation to introduce structural variation and reduce original language patterns.

Step 2 - Google Translation (EN → TR)

The output from Step 1 is translated into Turkish using Google Translate. This stage introduces additional syntactic distortion and cross-lingual variation based on rule-based and statistical translation differences.

Step 3 - DeepL Translation (Optional, TR → JA)

If a DeepL API key is provided, the Turkish text is further translated into Japanese using DeepL API. This optional step improves linguistic diversity and adds a second independent translation engine into the pipeline.

Step 4 - Final Reconstruction (DeepSeek) The resulting text is translated back into the original input language using the DeepSeek LLM. This step removes accumulated translation artifacts, restores readability, and reconstructs natural sentence flow while preserving semantic meaning.

image

Quick start(Windows/Linux/MacOS)

git clone https://github.com/kurelas/text-humanizer.git
cd text-humanizer
pip install -r requirements.txt

copy .\config\config.example.toml .\config.toml #config file creation

python main.py # run humanizer

Configuration

Edit config.toml to configurate text-humanizer:

Option Description
target_language Which language to use used for output/input(en, ja, zh, ko, de, fr, es)
deepseek_api_key Your DeepSeek API key
deepl_api_key Your DeepL API key(optional)
base_url Override API base url(empy = provider default)
model Model slug(empty = provider default)
temperature LLM temperature(1.3 recommended)

License

MIT License.

text-humanizer (kurelas/text-humanizer)

translation
Searching Database…

363 ·GitHub
Text-humanizer
What is this?text-humanizer is a free, open-source CLI tool that rewrites AI-generated text through a multilingual LLM-powered pipeline to make it read more naturally and bypass AI-content detectors like GPTZero and Turnitin.
Is this a library I import, or a standalone tool I run?It is a standalone command-line tool: you configure it through a config.toml file and run it directly, not a library you import into your own code.
How do I get it running?Edit config.toml with your target language and API keys, then run it; it works the same way on Windows, Linux and macOS.
Do I need an API key before I can use it?Yes, you need a DeepSeek API key for the core rewrite and reconstruction steps; a DeepL API key is optional and only unlocks an extra Turkish-to-Japanese translation pass.
Does it guarantee my text will pass AI detectors?No, the README only claims it 'bypasses most of' AI detectors, not all, and skips the extra DeepL translation pass entirely if you don't provide that key.
How does the humanization pipeline actually work?Text is first rewritten by DeepSeek with a Chinese intermediate translation, then machine-translated English-to-Turkish via Google Translate for extra syntactic distortion, optionally Turkish-to-Japanese via DeepL, and finally reconstructed back to the original language by DeepSeek to restore natural readability.
What languages can I humanize text in?The pipeline supports 8 languages (English, Japanese, Chinese, Korean, German, French, Spanish), selected via the targetlanguage option in config.toml.
Is the DeepL translation step required?No, it's optional. The Turkish-to-Japanese DeepL pass only runs if you supply a deeplapikey; otherwise the pipeline goes straight from Google Translate output to the final DeepSeek reconstruction.
Can I tweak how the rewriting behaves?Yes, config.toml also lets you override the LLM base URL, model slug, and temperature (1.3 is the recommended default).
What are similar projects?im-not-ai, Turnitout-Humanizer, spaCy, languagetool, Hy-MT2, LibreTranslate, jieba, newspaper
Which topics is it in?text-processing, natural-language-processing, translation, nlp, multilingual, deepseek, command-line-tool, free, api, chinese, tool, llm
Creator
kurelas
Size
0.6 MB
Stars
363
Updated
CommitOct 17
Age
12d
Commits · 90 Days