Humanizer-zh is a tool for removing AI-generated traces from Chinese text to make it read more naturally and authentically human-written.
The tool addresses the problem of AI-generated content bearing recognizable stylistic and linguistic markers. It identifies and helps revise twenty-four distinct patterns of AI writing organized into four categories: content patterns such as overemphasis on significance and legacy, language patterns including overuse of certain AI vocabulary and avoidance of copulas, style patterns like excessive dashes and bold formatting, and communication patterns including filler phrases and generic positive conclusions. The approach involves recognizing these markers and providing guidance for rewriting affected passages.
The tool suits editors and reviewers working with AI-generated content who want to improve its human quality, and those learning to recognize common AI writing patterns. It is designed for use within Claude Code through skill installation, supporting direct invocation and file processing workflows. The project is a Chinese adaptation of an existing humanizer tool, drawing on Wikipedia's guide to signs of AI writing and incorporating practical utilities from related projects.
The project maintains active development with regular updates to its skill definitions and documentation. The codebase includes structured pattern definitions and usage examples demonstrating application across marketing copy, academic abstracts, and blog content. Installation is supported through multiple methods including automated setup via npx and manual directory placement for different operating systems.
