Analysis updated 2026-05-18
Rewrite AI-generated text through a multi-stage translation pipeline to change its sentence structure.
Process text in any of eight supported languages using your own DeepSeek and optional DeepL API keys.
Study how round-trip translation and LLM rewriting can alter the statistical fingerprint of machine-written text.
| khrisat/text-humanizer | facebookresearch/vggt-omega | bytevisionlab/dreamlite | |
|---|---|---|---|
| Stars | 571 | 568 | 562 |
| Language | Python | Python | Python |
| Setup difficulty | — | hard | moderate |
| Complexity | — | 4/5 | 3/5 |
| Audience | general | researcher | researcher |
Figures from each repo's GitHub metadata at analysis time.
text-humanizer is a free, open source program that rewrites text produced by AI so that it reads more naturally. Its stated goal is to make AI-generated writing harder for detection services such as Turnitin and GPTZero to flag as machine-written, while keeping the original meaning. It works with eight languages, including English, Japanese, Chinese, Korean, German, French, and Spanish. The tool works as a pipeline, meaning the text passes through several stages in order. In the first stage, a large language model called DeepSeek rewrites the text into a version that means the same thing but uses different sentence structures and phrasing. As part of this step it also translates the content into Chinese as an intermediate form, which the author says introduces structural variation. In the second stage, the result is run through Google Translate into Turkish, which adds further changes in sentence structure from the differences between translation systems. There is an optional third stage that, if you supply a key for the DeepL translation service, translates the Turkish text into Japanese to add more variation from a second independent engine. In the final stage, DeepSeek translates everything back into the original language, cleaning up the artifacts left by the earlier translations and restoring readable, natural sentences. To use it, you clone the repository, install its requirements with Python's package installer, copy the example configuration file, and run the main script. It works on Windows, Linux, and macOS. Configuration is handled in a settings file where you set the target language, your DeepSeek key, an optional DeepL key, and a few model options such as the temperature, which controls how varied the rewriting is. The recommended temperature is listed as 1.3. The README is short and focused on what the tool does and how to set it up. It does require you to supply your own API keys for the underlying services. The project is released under the MIT License, which is a permissive open source license. The repository describes a single-purpose rewriting tool rather than a larger application.
An open source pipeline that rewrites AI-generated text through multiple translation passes to evade AI detection tools.
Mainly Python. The stack also includes Python, DeepSeek, Google Translate.
Mainly general.
This repo across BitVibe Labs
Verify against the repo before relying on details.