tt-humanizer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

TikTok Humanizer

TikTok 真人化工具

Rewrites a spoken script (and caption) to remove AI tells, and audits a finished draft against the 2026 TikTok checklist before you film. The problem this solves is specific to video: a script that reads fine on the page can sound robotic out loud. Written-not-spoken phrasing, perfect parallelism, and AI vocabulary all expose themselves the second a human says them to camera.

Based on Wikipedia's "Signs of AI writing" taxonomy plus TikTok-specific spoken patterns (the muted-first hook, the no-intro open, completion-rate structure).

重写口播脚本（及字幕）以去除AI痕迹，并在拍摄前根据2026年TikTok审核清单对完成的草稿进行审核。它解决的是视频特有的问题：在纸上读起来通顺的脚本，念出来却可能显得生硬机械。书面化而非口语化的表达、完美的平行句式以及AI专属词汇，一旦由真人对着镜头说出就会暴露无遗。

基于维基百科的“AI写作特征”分类体系，结合TikTok特有的口语表达模式（静音开场钩子、无介绍式开篇、完播率导向结构）。

When to use

使用场景

Before filming any AI-drafted spoken script (rewrite mode)
Pre-film review of a finished script + caption (audit mode, see
```
sub-skills/post-audit.md
```
)
When a script "reads fine but sounds off" when you say it out loud

AI生成的口播脚本拍摄前（重写模式）
完成的脚本+字幕拍摄前审核（审核模式，详见
```
sub-skills/post-audit.md
```
）
脚本“读起来通顺但念出来别扭”时

Input

输入内容

A spoken script (the hook line plus the body), optionally the caption, and optionally voice samples (the user's past scripts or how they actually talk).

口播脚本（包含钩子句和正文），可选提供字幕，还可提供语音样本（用户过往脚本或真实说话方式）。

Output

输出内容

Rewritten script that sounds spoken, not written
A diff showing what changed and why
Caption char count (flagging over 2,200) when a caption is included
Confidence: "human", "mixed", "AI-likely"

重写后的脚本，听起来像口语而非书面语
显示修改内容及原因的差异对比
若提供字幕，会统计字幕字符数（超过2200字符时发出提示）
置信度标签：“human（真人）”、“mixed（混合）”、“AI-likely（疑似AI生成）”

Modes

模式

bash

undefined

bash

undefined

Default: scrub AI tells and fix spoken-word issues

默认模式：清除AI痕迹并修复口语表达问题

tt-humanizer <script>

Forensic only - minimum touch, just kill model leakage

取证模式：最小改动，仅消除模型泄露痕迹

tt-humanizer --mode forensic <script>

Audit - detection-only pass-fail review, no rewrite

审核模式：仅检测，通过/不通过评审，不进行重写

Runs the 2026 TikTok pre-film checklist: first 1-3 second hook strength,

执行2026年TikTok拍摄前检查清单：开头1-3秒钩子吸引力、

muted-first text, completion design, caption fit, hashtag and settings sanity.

静音开场文本、完播率设计、字幕适配性、话题标签及设置合理性。

Returns Blockers + Warnings + suggested fixes. See sub-skills/post-audit.md.

返回阻塞问题+警告+建议修复方案。详见sub-skills/post-audit.md。

tt-humanizer --mode audit <script>

undefined

tt-humanizer --mode audit <script>

undefined

The three passes

三个处理阶段

Pass 1 - SCRUB (delete or replace)

第一阶段 - 清洗（删除或替换）

Apply the tiered catalogs in

references/scrub-rules.md

Forensic (always on): real model leakage no human says. AI tool markers (oaicite, contentReference, turn0search0), knowledge-cutoff disclaimers ("As of my last update"), template blanks ([Your Name]), and em dash overuse.
Strict (default on): bad spoken-word style regardless of origin. Vocabulary swaps (leverage -> use, delve -> look at, harness -> use, foster -> build), filler adverbs (fundamentally, essentially, ultimately), written connectives ("moreover", "furthermore"), dead filler ("hey guys", "without further ado"), and dead closers, both spoken ("thanks for watching", "don't forget to subscribe") and caption-level ("What do you think?", "Drop your thoughts below", bare "Let me know in the comments").

应用

references/scrub-rules.md

中的分层规则：

取证级（始终开启）：清除真人绝不会使用的模型泄露痕迹。包括AI工具标记（oaicite、contentReference、turn0search0）、知识截止声明（"As of my last update"）、模板占位符（[Your Name]）以及破折号滥用。
严格级（默认开启）：清除所有不良口语表达风格，无论来源。包括词汇替换（leverage→use、delve→look at、harness→use、foster→build）、填充副词（fundamentally、essentially、ultimately）、书面连接词（"moreover"、"furthermore"）、无效填充语（"hey guys"、"without further ado"），以及无效结束语（口语类："thanks for watching"、"don't forget to subscribe"；字幕类："What do you think?"、"Drop your thoughts below"、单纯的"Let me know in the comments"）。

Pass 2 - BREAK (make it sound spoken)

第二阶段 - 打破（使其口语化）

Replace full grammatical sentences with how a person actually talks: contractions and fragments. "It is something that you should consider" becomes "you should try this".
Break perfect parallel structures ("faster, cheaper, easier") with one asymmetric, specific line.
Vary line length. A teleprompter rhythm (every line the same length) sounds robotic out loud. Add a short punch line.
Read-aloud test: flag any line that needs two breaths or trips the tongue.

用真人实际说话的方式替换完整的语法句：使用缩合形式和断句。例如将“It is something that you should consider”改为“you should try this”。
用一句不对称的具体内容打破完美的平行结构（如“faster, cheaper, easier”）。
调整语句长度。提词器式的节奏（每句长度相同）念出来会显得生硬。添加简短的点睛语句。
朗读测试：标记任何需要换两次气或容易拗口的语句。

Pass 3 - ADD (human fingerprints)

第三阶段 - 添加（真人特征）

Require where the content allows:

1 specific number (replace "many", "a lot", "a few")
1 named entity (a real tool, person, or place)
1 first-person concrete detail ("the third take", "my 2am edit")
the spoken register: how this person would actually say it

If the input lacks these, ask the user for a number or detail. Do not fabricate.

在内容允许的情况下，添加以下元素：

1个具体数字（替换“many”、“a lot”、“a few”等模糊表述）
1个命名实体（真实工具、人物或地点）
1个第一人称具体细节（如“the third take”、“my 2am edit”）
口语语体：符合用户真实说话方式

若输入缺少这些元素，需向用户索要数字或细节，不得编造内容。

Non-negotiable rules

不可协商规则

Global voice rules: see root

SKILL.md

Voice rules. Additional skill-specific rules:

Scrubbing is always in scope. When asked to humanize, de-AI, finalize, or publish a script or caption, run at least the forensic + strict passes before it ships. This holds when the user wrote the draft themselves, says they love it as-is, or is in a hurry. Author identity, "it's already good," and time pressure are never reasons to skip the scrub. The forensic + strict pass changes no meaning and takes seconds: run it, then ship. If a constraint truly forbids touching the text, say so explicitly and name every tell left in; the default is to scrub, not to wave it through.
Preserve the user's actual claim and meaning. "Preserve their voice" covers voice quirks and what they are claiming, NOT corporate-speak, filler openers, or AI-tell phrasing. Stripping "leverage / fundamentally / in today's fast-paced world" is not changing their voice; it is the job.
Never introduce facts that were not in the input. If a number is missing, ask.
Keep it sayable. Every line has to survive being read out loud in one breath.
Keep the user's voice quirks (their slang, their pacing, lowercase texting style in the caption).

通用语音规则：详见根目录下的

SKILL.md

中的语音规则。额外的技能专属规则：

清洗始终是必做项。 当要求对脚本或字幕进行真人化、去AI化、定稿或发布时，至少要执行取证级+严格级清洗后再交付。即便脚本是用户自己写的、用户表示喜欢现有版本，或者时间紧张，也必须执行。作者身份、“已经很好了”以及时间压力绝不能成为跳过清洗的理由。取证级+严格级清洗不会改变原意，且仅需数秒：执行后再交付。若确实存在无法修改文本的限制，需明确说明，并列出所有遗留的AI痕迹；默认操作是清洗，而非直接放行。
保留用户的真实主张和原意。“保留用户语气”指的是保留用户的语音习惯和他们要表达的内容，而非官话、填充式开场白或AI特征句式。去除“leverage / fundamentally / in today's fast-paced world”这类表达并非改变用户语气，而是本职工作。
不得引入输入中没有的事实。若缺少数字，需向用户询问。
确保内容易于朗读。每句都必须能一口气读完。
保留用户的语音习惯（他们的俚语、语速、字幕中的小写短信式风格）。

TikTok-specific tells this skill catches

本技能识别的TikTok专属AI痕迹

A hook line that is written, not spoken ("In this video, I will demonstrate..").
A greeting or logo intro before the payoff ("hey guys, welcome back").
The spoken hook and the on-screen text saying the identical words.
A caption over 2,200 chars, or a 12-hashtag wall.
Perfect parallel tricolons read aloud ("learn, grow, succeed").
A "call to action" stacked five deep.
AI vocabulary that no one says on camera (leverage, utilize, robust, seamless).

书面化而非口语化的钩子句（如"In this video, I will demonstrate.."）。
核心内容前的问候或logo介绍（如"hey guys, welcome back"）。
口播钩子与屏幕文字完全一致。
字幕超过2200字符，或堆砌12个话题标签。
朗读完美的三并列句式（如"learn, grow, succeed"）。
堆叠五层的“行动号召”。
无人会在镜头前使用的AI专属词汇（leverage、utilize、robust、seamless）。

Example

示例

See

references/examples.md

for worked before/after rewrites of spoken scripts.

口播脚本重写的前后对比示例详见

references/examples.md

。

Files

文件说明

```
SKILL.md
```
- this file (rewrite scrubber + audit-mode entry)
```
references/scrub-rules.md
```
- vocabulary swaps and spoken-word fixes by tier
```
references/examples.md
```
- worked before/after script rewrites
```
references/audit-checklist.md
```
- the pre-film checklist with thresholds
```
sub-skills/post-audit.md
```
- pre-film audit workflow (detection-only, no rewrite)

```
SKILL.md
```
- 本文件（重写清洗器+审核模式入口）
```
references/scrub-rules.md
```
- 按层级划分的词汇替换及口语表达修复规则
```
references/examples.md
```
- 脚本重写的前后对比示例
```
references/audit-checklist.md
```
- 带阈值的拍摄前检查清单
```
sub-skills/post-audit.md
```
- 拍摄前审核流程（仅检测，不重写）

Voice profile mode (

--mode profile

)

语音档案模式（

--mode profile

）

tt-humanizer --mode profile

builds or updates the user's Voice & Brand Profile at

../../references/voice-profile.md

from 3-6 of their real TikTok posts pasted in (portable, no token) or, if a read token is set, from pulled activity. Once filled, every writing skill in this bundle drafts in the user's voice automatically. See

sub-skills/voice-profile.md

. Triggers: "build my voice profile", "learn my voice".

tt-humanizer --mode profile

可根据用户粘贴的3-6条真实TikTok帖子（可移植，无需令牌），或在设置读取令牌的情况下根据抓取的内容，构建或更新

../../references/voice-profile.md

中的用户语音与品牌档案。档案完成后，本套件中的所有写作技能都会自动以用户的语气生成内容。详见

sub-skills/voice-profile.md

。触发指令：“build my voice profile”、“learn my voice”。