npario/humanizer

TEXT GENERATIONPricing:Input $1.2 / Cached $0.24 / Output $4.8Concurrent Unit Cost:1Model Size:12BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

npario/humanizer is a 12 billion parameter model designed to rewrite AI-generated text into more human-like prose, supporting both English and Chinese. Developed by npario, it is specifically trained to preserve all factual details like numbers, dates, and names while improving natural language flow. The model achieves 95% human classification by Originality.ai and maintains high factual fidelity, making it ideal for refining AI drafts of emails, essays, and reports for a more authentic tone.

Loading preview...

humanizer: AI-to-Human Text Rewriter

humanizer is a 12 billion parameter model developed by npario that specializes in transforming AI-written drafts into text that reads as if a human wrote it. It supports both English and Chinese, making it versatile for a global audience. A key differentiator is its rigorous training to preserve all factual elements—numbers, units, dates, names, and quotes—without adding new information, ensuring accuracy while enhancing readability. The model is designed to run locally, offering privacy and offline functionality.

Key Capabilities

  • Human-like Rewriting: Rewrites AI-generated content (emails, essays, reports, forum posts) to sound natural and human.
  • Fact Preservation: Meticulously retains all factual details, including numbers, dates, names, and quotes, ensuring no information is lost or altered.
  • Multilingual Support: Handles both English and Chinese drafts effectively.
  • High AI Detector Evasion: Achieves 95% human classification by Originality.ai at its strictest setting, significantly outperforming previous versions and other tools.
  • Factual Fidelity: Maintains high accuracy, with 376 out of 420 English rewrites having no factual problems according to a strict LLM judge.
  • Optimized Quantization: Features advanced quantization techniques for smaller GGUF files (down to 3.9 GB) that maintain high agreement with the full bf16 model, enabling efficient local deployment on devices with limited memory (e.g., 8 GB RAM).
  • Local Operation: Available as a standalone app for Mac and Windows, or via llama.cpp, MLX, transformers, and vLLM for offline use.

Good for

  • Refining AI-generated content: Ideal for anyone needing to make AI drafts sound more authentic and less robotic.
  • Ensuring factual accuracy: Critical for professional documents where factual integrity is paramount.
  • Local and private processing: Suitable for users who require offline functionality or wish to keep their data private.
  • Developers integrating humanization: Provides clear usage instructions and prompt formats for integration into agents and applications.

Finetunes
2 models