humanize-text (lynote-ai)

Properties

category: anti-detector
maturity: growing
approach: Open-source pipeline that runs two LLM rewrites at temperature 1.3 followed by two hops across different neural machine-translation engines, publishing every intermediate step so the transformation is auditable rather than a black box.
access: open-source
owns: A transparent, self-hostable text-humanization pipeline offering four documented methodologies a user can read, modify, and run locally, rather than a single hidden rewriting model.
distinctFrom: Unlike undetectable-ai and stealth-writer (closed, hosted rewriting engines) or the direct-rewrite approach of blader/humanizer (an agent skill), humanize-text's core technique is round-tripping text through translation engines and multiple LLM passes with every step published, aimed at researchers or users who want to inspect and modify the transformation rather than trust a single proprietary model.
source: lynote-ai/humanize-text (GitHub)
notes: Quote is the GitHub API's own repository description field, confirmed verbatim via the API. MIT licensed, 3,037 stars, created 2026-05-18, actively pushed as of 2026-09-18 (within this lens's window). Predates the strict 2026-09-04 window by creation date; included as an untracked gap-fill given continued active development and star growth. Specific language/engine names (e.g. which two NMT engines, which intermediate languages) were reported only by a third-party summarization pass over the README and are not repeated here since they could not be independently reconfirmed verbatim in this session; the approach description above is limited to what the repository's own description field supports.

Provenance

Source: https://github.com/lynote-ai/humanize-text

Properties

category

anti-detector

maturity

growing

approach

Open-source pipeline that runs two LLM rewrites at temperature 1.3 followed by two hops across different neural machine-translation engines, publishing every intermediate step so the transformation is auditable rather than a black box.

access

open-source

owns

A transparent, self-hostable text-humanization pipeline offering four documented methodologies a user can read, modify, and run locally, rather than a single hidden rewriting model.

distinctFrom

Unlike undetectable-ai and stealth-writer (closed, hosted rewriting engines) or the direct-rewrite approach of blader/humanizer (an agent skill), humanize-text's core technique is round-tripping text through translation engines and multiple LLM passes with every step published, aimed at researchers or users who want to inspect and modify the transformation rather than trust a single proprietary model.

source

lynote-ai/humanize-text (GitHub)

notes

Quote is the GitHub API's own repository description field, confirmed verbatim via the API. MIT licensed, 3,037 stars, created 2026-05-18, actively pushed as of 2026-09-18 (within this lens's window). Predates the strict 2026-09-04 window by creation date; included as an untracked gap-fill given continued active development and star growth. Specific language/engine names (e.g. which two NMT engines, which intermediate languages) were reported only by a third-party summarization pass over the README and are not repeated here since they could not be independently reconfirmed verbatim in this session; the approach description above is limited to what the repository's own description field supports.