---

title: miaoCut sdk: docker app_port: 7860

MiaoCut — Free Online AI Background Remover

MiaoCut is a free, no-signup online AI image toolkit. Its core feature is one-click background removal — upload any photo and an AI segmentation model (BiRefNet) erases the background in about a second and returns a transparent PNG. It also ships a small family of related tools: ID photo maker, watermark remover, old-photo restoration, and format converters.

🔗 Remove backgrounds online with MiaoCut's free background remover — no account, no credit card, no watermark.

---

✨ Why MiaoCut

• 🚀 Instant — cutouts in ~1 second, no waiting room.
• 🎨 Full quality — exports a transparent PNG at the original resolution, no watermark.
• 🔒 Privacy-first — images are processed in memory and discarded immediately; nothing is stored, nothing is used to train AI.
• 🆓 Completely free — no hidden fees, no subscription traps, no signup.
• 🌍 Works anywhere — runs in any browser, on desktop or mobile.

🧰 Tools

Every tool is available in MiaoCut's free AI image toolkit:

Tool	Link	What it does
AI Background Remover	Free AI Background Remover	One-click background removal. Two modes: Sharp (fast, hard edges) and Fur (fine hair / fur / soft edges).
Product Photo Background Remover	/product-photo-background-remover/	Clean white-background and square-canvas product shots for e-commerce.
Portrait Background Remover	/portrait-background-remover/	Headshots, profile pictures, and social media cutouts.
ID / Passport Photo Maker	/id-photo-maker/	Auto cutout + background color swap + print layout for common ID photo specs.
Watermark Remover	/watermark-remover/	Brush over a watermark and let AI inpaint it away (LaMa).
Old Photo Restoration	/old-photo-restoration/	Repair fading, reduce noise, and enhance detail in old photos.
JPG → Transparent PNG	/jpg-to-transparent-png/	Turn a JPG into a true transparent PNG with a real alpha channel.
PNG → JPG (white background)	/png-to-jpg-white-background/	Flatten a transparent PNG onto white (or any color) and export as JPG.

📚 Guides: step-by-step tutorials on how to remove a background in PowerPoint, GIMP, and more.

🔬 How the cutout works

The background-removal pipeline picks one of two profiles per request:

• sharp (default, ~1s) — BiRefNet produces the mask, followed by gentle gamma correction and 1px anti-aliasing. Best for people, products, and logos with hard edges.
• fur (~3–5s) — adds alpha matting (closed-form solve on the trimap's unknown band) plus foreground-color estimation to remove background-color contamination on semi-transparent pixels. Best for white cats, curly hair, feathers, and other soft edges.

🏗️ Architecture

┌─────────────────────────────────┐      ┌────────────────────────────────────┐
│      Frontend (static)          │      │       Backend (API)                │
│  HTML + Tailwind CSS + vanilla  │ ───▶ │  FastAPI + BiRefNet (rembg)         │
│  JS, deployed on Cloudflare     │      │  on Hugging Face Docker Spaces ×3   │
│  Pages                          │      │  (miao_cut / miao_cut2 / miao_cut3) │
└─────────────────────────────────┘      └────────────────────────────────────┘

• Frontend — static HTML pages (index.html + compiled output.css) with EN/ZH/HI/ID/PT-BR/BN/FIL/UR i18n, client-side image compression, XHR upload with real progress. Served by Cloudflare Pages.
• Backend — FastAPI (single worker). AI models: BiRefNet-General-Lite (INT8 quantized) via rembg for cutouts, MediaPipe Face Mesh for ID-photo cropping, LaMa inpainting for watermark removal. Rate limiting (SlowAPI) and concurrency control live in process memory — no Redis or external state.

🚀 Run it yourself

Requirements: Python 3.11+ (backend), Node.js 18+ (frontend build only).

Backend

pip install -r requirements.txt

# Start the dev server. ⚠️ Use `python main.py`, NOT `uvicorn main:app`
# (the string form re-imports main.py and loads BiRefNet twice).
python main.py
# First run downloads the model weights to .u2net/. Serves on http://127.0.0.1:8000

Don't pass --reload (onnxruntime's thread pool deadlocks after fork) or --workers 2+ (rate-limit and concurrency budgets live in process memory).

Frontend

npm install
npm run build        # builds output.css + generates localized pages + sitemap
npm run watch:css    # watch mode while editing HTML

After changing any class in an HTML file, re-run npm run build:css and commit the updated output.css — production serves it as-is.

Docker

docker build -t miaocut-api .
docker run --rm -p 7860:7860 -e PORT=7860 -e MAX_CONCURRENCY=1 -v miaocut-data:/data miaocut-api

⚙️ Configuration (environment variables)

Variable	Default	Description
PORT	8000	Port to listen on
TRUST_PROXY	0	Set 1 to read the real client IP from X-Forwarded-For
ALLOWED_ORIGINS	empty (skips check)	Comma-separated allowed frontend origins
MAX_CONCURRENCY	1	Max concurrent inferences (key knob against OOM)
CUTOUT_PROFILE	sharp	Default cutout profile (sharp / fur)
DATA_DIR	.	Directory for persisted feedback data
ENABLE_DOCS	1	Set 0 to disable the /docs endpoint

🛡️ Privacy & security

• Images are processed purely in memory and destroyed immediately after.
• No uploaded image is stored.
• User data is never used to train AI models.
• Origin verification + per-IP rate limiting guard against abuse.

🤝 Contributing

Issues and pull requests are welcome. If MiaoCut is useful to you, a ⭐ on GitHub and a link back to the free AI background remover are hugely appreciated — they help more people find a free, privacy-respecting background remover.

📄 License

Released under the MIT License. You're free to use, modify, and self-host it. Attribution back to miaocut.app is appreciated but not required.

The free tools, the website front end, and the CPU background-removal backend (main.py) in this repository are MIT-licensed. The paid batch service behind miaocut.app (credits, subscriptions, the GPU pipeline, and its API gateway) is developed in a separate private repository and is not open source.

---