lynote-ai/humanize-text

Open-source pipeline and reference implementations for improving the readability and natural cadence of AI-assisted drafts.

What it solves

AI-generated text often carries statistical fingerprints that make it easily detectable by AI detectors and can feel unnatural in cadence. This toolkit provides a system to rewrite AI-assisted drafts to make them feel more natural and less likely to be flagged as machine-generated.

How it works

The project implements a multi-step pipeline that combines Large Language Models (LLMs) and Neural Machine Translation (NMT) engines to disrupt AI patterns. The recommended "Standard Pipeline" follows a 4-step chain:

  1. LLM Rewriting (Chinese): The text is rewritten and translated into Chinese using an LLM (e.g., DeepSeek) at a high temperature (1.3) to introduce creative variation.
  2. LLM Rewriting (Japanese): The Chinese text is further rewritten and translated into Japanese, using the first step as conversation history for coherence.
  3. NMT Translation (Finnish): The text is routed through Google Translate to Finnish to introduce structural disruption via a linguistically distant language.
  4. NMT Translation (English): The text is finally translated back to English using the Niutrans engine to reconstruct the text without the original AI fingerprint.

Who it’s for

Developers and writers who want to improve the readability and natural flow of AI-assisted drafts while maintaining the original meaning and style.

Highlights

  • Multi-Engine Approach: Combines LLMs (DeepSeek, OpenRouter) with multiple NMT engines (Google, Niutrans) to ensure no single-engine fingerprint survives.
  • Linguistic Distance: Uses a chain of distant languages (Chinese $\rightarrow$ Japanese $\rightarrow$ Finnish) to maximize structural restructuring.
  • Reference Implementations: Includes four distinct humanization methodologies (translation chaining, multi-turn rewriting, detection-guided feedback, and mixed-engine translation).
  • Integration Ready: Provides both a Python CLI and an n8n workflow for no-code automation.

Related

  • Project
    fromleda/text-humanizer

    An open-source pipeline that humanizes AI-generated text by using a multi-step rewriting and cross-lingual translation process to bypass AI detectors.

  • Project
    larashero3-dotcom/lieflat-less-ai-tone

    A linguistic research project and toolset that quantifies "AI flavor" in Chinese text and provides a set of rules to remove these markers for more human-like writing.

  • Project
    KKKKhazix/human-writing

    A framework and toolset for AI agents to eliminate "AI-style" Chinese writing by enforcing human-like rhythms, factual substance, and structural checks.

  • Dispatch
    Detecting LLM-Generated Text with Classical Machine Learning – Results and Insights

    A scikit-learn TF‑IDF + LinearSVC pipeline can distinguish modern LLM‑generated Chinese text from human‑written text with ~85% sentence‑level accuracy, demonstrating that “classical” ML still works for AI‑text detection.

  • Project
    LifelongLazyLearner/qu-ai-wei

    A Simplified Chinese rewriting skill that removes stiff, formulaic AI-generated patterns to make text sound more natural while preserving original facts and tone.