Apache-2.0llmRuns locally12B
humanizer: Local 12B Style-Rewriting Model
Local 12B model that rewrites drafts for a more natural reading style (English and Chinese).
- License
- Apache-2.0
- Parameters
- 12B
- HF likes
- 1,337 (2026-10-13)
- First seen
- 2026-09-11
TL;DR. humanizer is a llm model from jialinyyzz. Parameters: 12B.License: Apache-2.0. It is listed under Undetectable.ai.
| Developer | jialinyyzz |
|---|---|
| Type | llm |
| Architecture | Text-completion fine-tune of google/gemma-4-12B (not a chat model, per the card) |
| Parameters | 12B |
| Modalities | text |
| Languages | English, Chinese |
| License | Apache-2.0 — Apache-2.0 as listed on the Hugging Face card and GitHub LICENSE as of 2026-10-13. Not legal advice. |
| Hardware | Not yet verified — README lists self-reported peak memory on a Mac M5 Max (Q8_0 about 13.7 GB while rewriting). Not independently tested. (not independently tested) |
| Hugging Face likes | 1,337 (as of 2026-10-13) |
| Model card | https://huggingface.co/jialinyyzz/humanizer |
| First seen | 2026-09-11 |
humanizer is a 12B open-weight rewriting model. The identification URL is jialinyyzz/humanizer. The companion code and desktop app live at sgaofen/humanizer-local-model (the GitHub API also accepts the older sgaofen/humanize-model path).
This page is a spoke from Undetectable.ai alternatives. That hub is an editorial grouping for local style/readability rewriting. We do not promise that rewritten text will pass any detector, and we do not offer academic-cheating advice.
What we verified (2026-10-13)
| Field | Value | Source |
|---|---|---|
| Hugging Face likes | 1,337 | HF API jialinyyzz/humanizer |
| Card created | 2026-09-11 | HF createdAt |
| License | Apache-2.0 | HF cardData.license; GitHub license.spdx_id |
| Parameters | 12B (fine-tune of google/gemma-4-12B) |
Model card |
| Languages | English and Chinese | Model card |
| GitHub stars / forks | 198 / 31 | GitHub API sgaofen/humanizer-local-model |
| App | 0.3.2 for Mac (Apple silicon) and Windows x64 | HF card and GitHub README |
| Runtime without the app | llama.cpp (humanizer-12b-Q8_0.gguf recommended) |
Same |
The card states it is a text-completion model, not a chat model, and that chat templates should not be used.
How we describe it
Upstream marketing talks about drafts that “read like a person wrote them.” On forkalt that is framed as style and readability rewriting: vary sentence length, cut filler, keep facts. If you need a tool that claims to fool detectors, this site will not rank or endorse that use.
What we did not invent
- We have not timed the app on our hardware (the README cites ~3.6 s for a hundred-word email on an M5 Max; that is upstream, not a forkalt test).
- We have not reproduced fact-judge or detector tables from the card.
- A Linux desktop installer is not yet verified; Linux use is documented via llama.cpp.
- We have not audited training data.
See also writing & office and models.
Deploy options
FAQ
What does humanizer actually do?
The official card describes a 12B local model that rewrites AI-written drafts (emails, essays, reports, forum posts; English and Chinese) so they read more like a person wrote them, while keeping numbers, units, dates, names and quotes. forkalt treats that as style and readability rewriting, not as a detector-bypass product.
Is there a desktop app?
Yes. The Hugging Face card and GitHub README (checked 2026-10-13) point to Humanizer 0.3.2 for Mac with Apple silicon (.dmg) and Windows x64 (installer or portable zip) from GitHub Releases. First run downloads a GGUF from Hugging Face; after that the README says it works offline. We have not run the installer ourselves.
Will this make text “undetectable” or help with academic cheating?
We do not claim that, and we do not give advice on evading AI detectors or academic-integrity rules. The project publishes its own detector measurements; those are upstream self-reports, detectors change, and they are not a reason to use the model. Use it only on text you are allowed to edit.
What is the license?
Apache-2.0 on the Hugging Face card and the GitHub repository as of 2026-10-13. Not legal advice.
Sources & verification
- https://huggingface.co/jialinyyzz/humanizer
- https://github.com/sgaofen/humanizer-local-model
- https://huggingface.co/jialinyyzz/humanizer/blob/main/assets/app-en.png
License notes are not legal advice. Read the original LICENSE. Trademark & disclaimer.