Humanizer 12B — local model that rewrites AI drafts to read human
A 12B model (EN+ZH) that rewrites AI-written drafts so they read like a person wrote them while preserving every number, date, name, and quote; ships with Mac/Windows apps and runs fully offline.
Why it matters
Evaluated like a product, not a checkpoint: 95% of rewrites judged human by Originality.ai at its strictest setting, with a strict LLM fact judge reporting 376/420 rewrites with zero factual slips — plus a 2-bit GGUF down to 3.9 GB.
What you could build with it
A freelance writer or content team could run the 4-bit quant on a laptop to pass client drafts through a local de-AI-washing step, keeping sensitive text off cloud services while satisfying editors who reject obviously machine-written copy.
Does it hold up?
Strongest evidence of the batch: the author publishes per-quant fact-check and AI-detector numbers, a before/after gallery, and one-click Mac/Windows apps — but all evaluation is self-run, so independent verification is still missing.
Built with Humanizer 12B
- Trending on Hugging Face, 575 likes within ~7 hours of updatehuggingface · Strong early traction signal for a niche release; the model card includes before/after examples, a full evaluation card, and usage guides in English and Chinese.
- sgaofen/humanizer-local-modelgithub · 61 ★ · pipx-installable CLI ('hz') that rewrites whole documents and folders with the humanizer model, preserving headings, code, tables, and links.
- jialinyyzz/humanizer-GGUFgithub · 7 ★ · Official GGUF releases of every quant (2-bit through Q8_0), including quantization-aware-trained 2-bit/Q3/Q4 files with measured KL and fact-check results.
Learn more
First spotted on hf: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.