humanizer by jialinyyzz

humanizer by jialinyyzz

Rewrites AI-written drafts such as emails, essays, reports and forum posts so they read like a person wrote them, in English and Chinese. A 12B model that keeps numbers and names and runs locally.

Made by jialinyyzz on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: humanizer is a 12B model by jialinyyzz, based on Gemma 4 12B, that rewrites AI-written drafts so they sound like a person wrote them. It handles emails, essays, reports and forum posts in English and Chinese. It was trained not to alter figures, measurements, dates, names or quoted words, and not to add anything of its own. It runs on your own machine.

Who it suits: People who start from an AI draft and want it reworded in a natural voice without changing the facts.

Industry | Who uses it | How they use it | Example
Office work | Employees and managers | Reword AI-drafted emails and reports | A work email rewritten in a natural tone
Education | Students writing essays | Reword a draft in their own style | An essay draft rewritten
Online communities | Forum posters | Reword answers and posts | A forum answer or a Zhihu answer
Chinese-language writing | Chinese writers | Reword Chinese drafts | A Chinese work email rewritten

If you write everything by hand, or need a tool that checks facts for you, you can skip it.

Where to use it:

• A Mac app for Apple silicon and a Windows app, which work offline
• llama.cpp and other runtimes that load GGUF files
• Coding agents, through the AGENTS.md file the card points to

Use case | Who | What they get back
Work emails | Office staff | A rewrite with the same numbers, dates and names
Essays and reports | Students and analysts | A reworded draft that reads like a person wrote it
Forum posts | Community members | A reworded answer in English or Chinese

How to set it up:

• Windows, macOS or Linux with Python to fetch the files
• About 6 GB of disk space at 4-bit, or about 23.9 GB in BF16 (estimates)
• Optional: the maker's Mac or Windows app, so you do not run the model yourself

1. Install the Hugging Face hub tools.
2. Download the model files.
3. Load a GGUF file in llama.cpp or another runtime, or use the maker's app.

pip install -U huggingface_hub
hf download jialinyyzz/humanizer

How to use it: Open the app and paste a draft. The rewrite appears beside it with the new wording highlighted, and the app works offline. Check the output yourself before sending, with a close eye on figures, dates and names.

Example: You paste a stiff AI-written work email that mentions a date and an amount. The model returns a version in a more natural tone with the same date, amount and names, and nothing added.

For newcomers and experts: If you are new to this, install the app, paste a short email draft and compare the two versions side by side. Check every number and name before you use the rewrite.

Under the hood it is a 12B model, based on google/gemma-4-12B, in Safetensors and GGUF. The card says it was trained with no AI detector in the loop. Some of its smaller quantized files got extra training to stay close to the full model. At 3.9 GB the card reports 87 percent agreement with the full model's next token, against 70 percent for a standard build.

You paste an AI-written draft
The model rewrites it in a natural voice
Numbers, dates, names and quotes stay as they were
You review the highlighted changes

Good to know:

• Fact errors: the card's judge flagged 44 of 420 English rewrites, so read each result before sending it
• Detector result: the card reports 11 of 210 English rewrites flagged as AI by Originality.ai

Title: English rewrites flagged as AI by Originality.ai, out of 210
This release: 11 flagged
Previous release: 26 flagged

• License: apache-2.0
• Cost to run: no API key or paid service is named, only your own computer
• Data it reaches: the app and model run offline on your files and text, with the download as the only web step
• Not for: skipping a final read, since the card says to check numbers, dates and names yourself

On Hugging Face, 13 November 2026

• Task: text generation
• Library: gguf
• Parameters: 12B
• Weights: about 23.9 GB in BF16; about 6 GB at 4-bit (estimate)
• Formats: Safetensors, GGUF
• Downloads in the last 30 days: 391,936
• Likes: 2,260
• License: apache-2.0
• Languages: en, zh
• Based on: google/gemma-4-12B
• Quantized versions: 15
• Fine-tunes: 4
• Tags: gguf, safetensors, gemma4_unified, image-text-to-text, humanizer, transformers, text-rewriting, rewriting, paraphrase, style-transfer, llama.cpp, gemma4

Links

• Model: https://huggingface.co/jialinyyzz/humanizer
• Files: https://huggingface.co/jialinyyzz/humanizer/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download jialinyyzz/humanizer

Then follow the usage in its model card. Nothing in this kit runs by itself.