Humanize Text
lynote-ai/humanize-textReviewed by AI
Checked by RepoPicks Radar on
What is Humanize Text?
Humanize Text takes a draft written by AI and rewrites it so it sounds more like a person. It sends the text through four steps: two rewrites by a language model (DeepSeek by default) and two machine translations in different engines, hopping through Chinese, Japanese and Finnish before coming back to English. You run it as a Python command, as a Docker service with a REST endpoint, or as an n8n workflow, and you need your own model and translation keys. Every intermediate step can be printed, so you can see what changed.
How easy is it to install Humanize Text?
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
The same four steps on all three systems, with Linux and macOS matching the commands in the README exactly: install Python 3.10 or newer, clone the repository, run pip install, copy the example config and paste two API keys, then run one command per text. Docker is offered as an alternative for a REST endpoint. There is no installer, no window to click in, and nothing to download from the releases page.
Humanize Text on GitHub
- Stars
- 3,497on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +16stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python
- To build from source
- Python, Docker, Graphics card
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- DeepSeek or another OpenAI-compatible model provider (OpenRouter, Atlas Cloud, OrcaRouter, LiteLLM), Niutrans (translation), Google Translate (through deep-translator)
- Cost
- Free, API costs extra · The code is free (MIT), but it only works with paid keys: a language model provider (DeepSeek by default, or OpenRouter and others) and a Niutrans translation key. The maker's hosted service is sold separately at "1 credit per input word".
- License
- MIT (Permissive)
- Can replace
- Undetectable AI, QuillBot
- Last change on GitHub
- Oct 22, 2026
Humanize Text risk signals (4)
- Attentionmade to pass ai detectors
The project's purpose is to make AI text read as human-written and get past AI detectors: the pipeline aims to "break the statistical patterns typical of AI text" so "no single engine's fingerprint survives", the `focus` tier adds a "detection-guided feedback loop" for "Maximum restructuring", and the showcase table lists a "Detector verdict on our test" of "human" with confidence up to 0.9997. Topics include "undetectable-ai-alternative". Schools, journals and many employers have rules about using and disclosing AI writing; the README itself says "follow your institution's policies on AI use and disclosure. Please don't use it to misrepresent authorship."
AI review
- Infotext sent to several outside services
One run sends your draft through up to four outside services: "Step 1 LLM ... Step 2 LLM ... Step 3 Google Translate ... Step 4 Niutrans". Whatever you paste, including unpublished work, passes through the model provider and both translation engines.
AI review
- Infopromotes the makers paid service
The repository belongs to the company that sells the hosted version, and the README points to it repeatedly with tracking links, for example "Try it online" to "lynote.ai/ai-humanizer?utm_source=github&utm_medium=readme...", plus a comparison table "This repo | Lynote.ai" where the paid service uses "our own model". The README also says of the open training pipeline: "Model weights and full datasets are not included."
AI review
- Infoquality numbers from the maker
The scores in the README are the maker's own: "Expert evaluation on 50 text pairs" with an "Overall 9.1" out of 10 and "Key information retention: 100% (50/50 pairs)", and the detector results are "illustrative snapshots from one detector at one point in time, not a guarantee".
AI review
Evidence is shown in its original language.
What Radar could not check about Humanize Text
- Whether the rewritten text passes the detector your school, journal or client actually uses
- How much the model and translation keys cost for a long document
- How much meaning, citations and technical terms drift after four steps, which the README warns about
- Whether the Docker and n8n paths work as described
- What the model provider and the translation engines keep of the text you send