Humanize Text — AI Text Humanizer & Detector

Overview

Humanize Text is a self-hosted FastAPI application with two jobs: rewriting machine-sounding text so it reads more naturally, and estimating how likely a passage is to be AI-generated.

How it works

The Humanize mode offers three interchangeable methods:

• LLM rewrite: a pluggable LLM client (DeepSeek, OpenRouter or another compatible provider) varies sentence rhythm and vocabulary.
• Translation chain: a round trip through several languages using free translation engines, with no API key required.
• Mixed engine: blends several translation engines.

A post-processing stage cleans up the output.

The Detect mode runs a dependency-free ensemble of stylometric signals. It returns an AI-likelihood score, a confidence level and a per-signal breakdown such as formal connector density, so users can see why a passage was flagged. The codebase also contains detector modules based on Binoculars, RoBERTa and statistical methods.

Engineering notes

The API validates input (up to 20,000 characters), enforces request timeouts and maps errors to clean messages without stack traces. Interactive Swagger docs are included. A dark single-page UI offers mode switching, a live word count, copy-to-clipboard and keyboard shortcuts, with no frontend build step. The project ships with Docker Compose and a container healthcheck, a CI workflow, tests, documentation, and an n8n workflow for using the humanizer inside automations.