Humanize Chinese Text and Reduce AI Detection

Humanize Chinese detects AI-like patterns in Chinese writing and rewrites it to feel less synthetic - reducing AIGC (AI-generated content) signals in academic prose, or converting text into a specific style. It triggers on requests like 去AI味, 降AIGC, 去除AI痕迹, 让文字更自然, 改成人话, or 降低AI率, and follows a three-step core workflow: detect first, flagging rigid enumeration structures, mechanical connectors such as 综上所述, 值得注意的是, and 由此可见, abstract low-information wording, and repetitive sentence rhythm before touching anything; rewrite in the smallest useful pass, removing formulaic connectors and varying rhythm rather than regenerating the whole text, while keeping claims, facts, citations, and terminology intact; then validate that the result still says the same thing, reads less templated, and has not drifted from the facts or the target audience's register.

Use it when Chinese text needs to be checked for AI-writing patterns or rewritten to sound more natural, including academic-paper-specific AIGC reduction aimed at CNKI, VIP, or Wanfang-style detectors, or when the user wants the text converted into a specific style - casual, zhihu, xiaohongshu, wechat, academic, literary, or weibo. Style conversion is meant to happen only after the base text already reads naturally, changing tone, structure, and surface wording while keeping the meaning stable. For academic text specifically, it preserves a scholarly tone rather than over-casualizing, and it will not invent citations, evidence, or data to make a conclusion sound more grounded.

Input: Chinese text - a paragraph, paper, or report - and, optionally, a target style. Output: the main AI-like patterns found, a 1-3 bullet rewrite strategy, and the rewritten Chinese text, plus an optional note on any remaining weak spots. Concrete academic-phrase substitutions it favors include softening 本文旨在 to 本文尝试 or 本研究关注, 具有重要意义 to 值得关注 or 有一定参考价值, and 研究表明 to 前人研究发现 or 已有文献显示 - measured hedging over absolute certainty, without inventing new evidence.

python3 scripts/detect_cn.py text.txt -v
python3 scripts/compare_cn.py text.txt -a -o clean.txt
python3 scripts/academic_cn.py paper.txt -o clean.txt --compare
python3 scripts/style_cn.py text.txt --style xiaohongshu -o out.txt

When that local CLI toolkit is available, it runs detection, rewrite or compare, a re-detection pass on the cleaned file, then an optional style conversion; without it, the same detect-rewrite-validate sequence is done manually - deleting weak transitions, collapsing repetitive phrases, splitting overly symmetrical sentences at natural turns, and merging choppy ones.

Adapted from the voidborne-d/humanize-chinese open-source project and its CLI/script workflow for Chinese AI-text detection and rewriting.

Writers, students, and researchers working in Chinese who need AI-generated or AI-detector-flagged text rewritten to read naturally - especially academic authors preparing text to pass CNKI, VIP, or Wanfang-style AIGC checks without altering the underlying facts or citations.