First AI product · Free
Make ID photos with AI, right in your browser
Upload a portrait: the app detects the face, removes the background, straightens and crops to a standard ID-photo size, switches to a white, blue or custom background and sharpens the result. Photos are processed on your device and never uploaded.
1. Portrait photo
2. Size
3. Background colour
4. Fine-tune
Print sheet
Your result appears here. The first run downloads the AI models (about 37 MB), which the browser then caches.
Processing pipeline
Six steps, all running as JavaScript and WebAssembly on your device.
- Face detectionMediaPipe Face Landmarker: 478 landmarks; eye centres and chin are used.
- Background removalMODNet (ONNX Runtime Web) predicts an alpha matte; edges refined with a guided filter.
- Straighten and cropRotates along the eye line and places crown and chin according to the head-size ratio.
- New backgroundForeground colour estimation (blur-fusion) removes old-background fringes before compositing on the chosen colour.
- SharpeningUnsharp mask with a suggested amount based on sharpness measured on the face region.
- Export300 DPI PNG or JPG, or a 10×15 cm / A4 sheet with multiple copies.
Supported sizes
| Size | Pixels @300 DPI | Default head size |
|---|---|---|
| 2×3 cm | 236×354 | 70% |
| 3×4 cm | 354×472 | 70% |
| 4×6 cm | 472×709 | 72% |
| 3.5×4.5 cm | 413×531 | 75% |
| 2×2 inch | 600×600 | 60% |
ID photo requirements differ by receiving authority; head size and custom dimensions can be adjusted in the app.
Privacy and requirements
- Photos are processed entirely in the browser; no photo is uploaded to a server.
- The first run downloads about 37 MB (models + runtime); afterwards the browser cache is used.
- Requires a browser with WebAssembly (recent Chrome, Edge, Firefox or Safari).
- Front-facing, well-lit, single person; the larger the face in the source photo, the sharper the ID photo.
Technology and credits
- MediaPipe Face Landmarker — Google AI Edge, Apache-2.0.
- MODNet — Ke et al., “MODNet: Real-Time Trimap-Free Portrait Matting via Objective Decomposition”, AAAI 2022, Apache-2.0.
- ONNX Runtime Web — Microsoft, MIT.
- Guided filter — He, Sun, Tang, “Guided Image Filtering”, IEEE TPAMI 35(6), 2013.
- Blur-fusion foreground estimation — M. Forte, “Approximate Fast Foreground Colour Estimation”, ICIP 2021.
- How the tool was built: measurements and bugs
Need it inside your own system?
Shopsmon takes on customisation, integration and custom browser-based AI apps.