Rewrite AI text to sound human. Based on the Humanizer writing principles.
Built by Ji Huang.
- Model choice —
deepseek-v4-flash(fast, affordable) ordeepseek-v4-pro(highest quality). Thinking mode is disabled — rewriting works better with direct output. - Rewrite strength — Light (fix only obvious AI tells), Medium (balanced corrections), or Strong (full human voice).
- Preserve Markdown — Keep headings, lists, code blocks, and formatting intact through the rewrite.
- Full analysis mode — For power users: see the draft, an anti-AI audit of remaining tells, and the final version, so you understand what changed and why.
The app sends your text to the DeepSeek API with a detailed system prompt covering 29 AI writing patterns — from inflated vocabulary and emoji overuse to formulaic structures and soulless tone. The model rewrites the text and returns a natural, human-sounding version.
Thinking mode is disabled (thinking.type = "disabled"). DeepSeek's reasoning step consumes extra tokens but adds no value for a rewrite task. Prompt caching is automatic — the static system prompt hits cache on every call, so only the user's input text costs tokens.
source .venv/bin/activate
uv pip install -r requirements.txt
streamlit run app.pyCreate .streamlit/secrets.toml:
APP_PASSWORD = "your-password"
DEEPSEEK_API_KEY = "sk-your-api-key"- Push this repo to GitHub.
- Go to share.streamlit.io and sign in.
- Click New app, then select this repository, branch (
main), and main file path (app.py). - In the app dashboard, go to Settings → Secrets and add:
APP_PASSWORD = "your-password"
DEEPSEEK_API_KEY = "sk-your-api-key"- Click Save — the app redeploys with your secrets. It's now live at
https://<your-app>.streamlit.app.