Home › Blog › AI passport photos and privacy
AI passport photos and your privacy: what happens to your face
AI now does the parts of passport photography that used to need a studio: cutting you out of the background, cleaning up a soft face, framing you to the millimetre. But an AI passport photo raises a fair question — where does your face go while the AI works? The answer depends entirely on where the model runs, and that is the one thing most services don't make easy to check.
Can ChatGPT make a passport photo?
ChatGPT and similar chat assistants can edit images, but a chat model is the wrong tool for a compliant passport photo. Passport offices require exact physical dimensions (51 × 51 mm for a US passport, 35 × 45 mm for a UK one), a defined head height, 300 DPI print quality, and a photo that still shows what you actually look like. A generative model that redraws your image can subtly change facial features — the opposite of what an identity document needs, and in some countries an altered photo is grounds for rejection outright. What general AI chatbots are genuinely useful for is summarising your country's published photo rules so you know what to aim for before you open a dedicated tool.
The two kinds of AI photo tool
Every "AI passport photo" service does its machine learning in one of two places, and the difference decides your privacy:
- Server-side processing. Your photo is uploaded to the company's servers, the AI models run there, and the result is sent back. The company — and whichever cloud provider hosts it — possesses your portrait at least for the duration of processing, and their privacy policy governs what happens afterwards. This is how nearly all mobile app-store "passport photo" apps work.
- On-device processing. The AI model is downloaded once into your browser and runs on your own hardware. The photo never leaves your device, because there is no upload step in the pipeline at all. The tool's servers see only an anonymous browser session, not your face.
How Passport Room's AI works
Passport Room takes the on-device route. The background removal runs a MODNet segmentation model in your browser through onnxruntime-web — using your device's WebGPU when available, with a WebAssembly fallback on machines without it. The face enhance pass runs a GPEN-BFR-256 restoration model the same way, improving only the detected face region so the result stays recognisable. Both model files are downloaded once and cached, after which the tools keep working even with your network switched off. The photo is read straight from your file picker into the page's canvas; it is never sent anywhere.
Two honest limits of that approach: browser AI is lighter than data-centre AI, so very fine hair against a busy background cuts out less cleanly, and a heavily blurred source photo stays soft after enhancement. On-device privacy buys you those trade-offs; a server tool makes the opposite trade. Which trade is right for an identity photo is a privacy decision only you can make.
What AI passport photo tools can and can't do
| Task | AI handles it? | Reality check |
|---|---|---|
| Background removal and replacement | Yes, reliably | The core use case; a plain wall still beats a busy room for edge quality |
| Cropping to exact size and head height | Yes | Deterministic maths, done best by presets and guides rather than generation |
| Sharpening a soft face | Yes, within limits | Improves the face region; can't recover detail a camera never captured |
| Changing your clothes | Overlay only | Garment overlays are placed by you; generated bodies are not ID-safe |
| Generating a realistic new face | Yes — don't | A generated face is not your face; using one on an identity document is fraud, not editing |
Questions to ask any AI photo tool
- Where does processing happen — in my browser or on your servers?
- Is my photo stored after processing, and for how long?
- Is the model generative (does it redraw me) or analytical (does it cut and correct me)?
- Do exports carry a watermark, or is that behind a paywall?
- Does the tool work offline after setup?
Passport Room's answers, for comparison: processing happens in your browser; the photo is not uploaded or stored on any server; the models are analytical — they separate and correct your real photo rather than generating a new one; exports are free, unlimited and watermark-free; and the tools work offline once the models are cached. The feature list documents each of these in detail, and the privacy policy covers the account and cookie side.
FAQ
Is an AI passport photo accepted by passport offices?
Offices judge the result, not the tools used to make it. A photo that meets the size, head-height, background and quality rules is acceptable whether it came from a studio, a booth or a well-made home photo — and unacceptable if it doesn't. AI editing is fine as long as it does not alter your facial features.
Do AI passport photo apps store my photo?
Server-side apps usually process and sometimes store your photo remotely; their privacy policy is the only place this is stated honestly. On-device tools like Passport Room have nothing to store server-side because the photo never arrives there.
Why is my AI background removal rough around the hair?
Hair against a matching or busy background is the hardest segmentation case for any model. Shooting against a plain wall with even light gives the model the cleanest possible edge; after that, the manual adjustment panel lets you polish the result.
Can AI make my photo look more professional?
Yes — within ID rules. Background replacement, exposure correction and gentle face enhancement are all standard. What it should never do is change your features: the photo's job is to prove what you look like.
Where to go next
Try the pipeline yourself in the free passport photo maker, see what is being improved next in the AI roadmap, or learn the manual editor that sits alongside the AI passes.