Home › AI roadmap

The Passport Room AI roadmap

Passport Room's direction is simple to state: an AI-powered professional photo studio that anyone can open in a browser, that never asks for your face to be uploaded, and that keeps getting closer to a human photographer's judgement. This page separates what exists today from what is planned and what is still experimental — no promises dressed up as features.

Available today

These features are live in the photo maker right now:

• On-device background removal. A MODNet segmentation model runs in your browser through onnxruntime-web with WebGPU acceleration and a WebAssembly fallback. The photo never leaves your device, and after the one-time model download the tool works offline.
• AI face enhance. A GPEN-BFR-256 restoration model improves the detected face region only — sharpening a soft indoor photo while keeping you recognisable — with a gentle "Natural" preset and a stronger preset for rougher sources.
• Exact size engine. Country presets rendered at exact millimetre dimensions and head-height ratios at 300 DPI, with custom mm/inch sizes supported.
• Print-sheet composition. True-scale tiled sheets with cutting guides for home printers and photo kiosks.
• Outfit try-on. A manual garment overlay editor — five bundled formal garments plus your own PNGs, composited on a canvas in your browser.
• Manual adjustment panel. Brightness, contrast, saturation, exposure, shadows, highlights, sharpness, clarity and noise, applied on-device and resettable at any time.

Planned improvements

Work in progress or firmly on the list — these describe direction, not dates, and each one ships only when it meets the privacy bar of running on-device:

• Better edges on difficult hair. The current segmentation struggles most with fine hair against a busy background. The plan is a stronger segmentation model and edge-refinement tooling, so a cut-out against a cluttered room looks as clean as one shot against a studio wall.
• Natural photo enhancement. A broader, passport-aware enhancement pass beyond the face region — evening out exposure across the shoulders, cleaning background colour noise, and keeping skin tones true rather than "beautified", because identity photos must stay honest.
• Identity preservation first. A principle, not a feature: any enhancement must never change facial geometry. As stronger generative models appear, the project's filter is that the person in the exported photo must remain the person who took it.
• More clothing options. Expanding the outfit library beyond the five bundled garments, and improving the placement tools (auto-fit guides, better colour matching at the collar) so the overlay looks natural without manual fiddling.
• Intelligent lighting correction. Detecting uneven or coloured light in the source photo and correcting it automatically — matching the even, neutral lighting a studio would have produced — instead of leaving it entirely to manual sliders.
• Photo quality pre-checks. Warnings before you export: head height outside the guides, shadows on the background, low resolution for the chosen preset, or a source photo that is too small to meet the standard.
• Passport-photo compliance checks. Automated checks against the published rules of each supported country — head position, background uniformity, expression cues — as an advisory layer. The final judgement always belongs to the issuing office, and the tool will say so.
• Faster processing. Reducing the time from photo selection to preview — quicker model loading, lighter WebAssembly paths for older devices, and caching improvements so repeat photos process near-instantly.

Experimental

Being explored, not promised. These may change shape, ship differently, or not ship at all:

• Automatic pose and head-angle guidance. Live feedback while you take the source photo — telling you to raise the phone, square your shoulders, or move nearer a window, so the source photo is right before any editing begins.
• Smarter form matching. Choosing the correct preset from the name of the application form you are filling in, so the guessing step between "visa" and "passport" sizes disappears.
• Multi-photo family sessions. Processing several family members in one sitting with shared settings, then producing a combined print sheet.

The principles behind the roadmap

1. On-device by default. If a capability cannot run on your device, it does not ship until it can — or it ships server-side only if privacy can be guaranteed and stated plainly. This is why the tool works offline and why your photo has no upload step.
2. Honesty over capability. Features are documented with their limits. No "guaranteed acceptance" claims, because no software can promise what a human clerk will decide.
3. Free access. New AI features follow the free-forever model — no watermarks, no export paywalls.
4. Identity-safe enhancement. The AI improves the photo you took; it never invents the person in it.

How this page is maintained

This roadmap is updated as features land. When something moves from planned to available, it is added to the feature list with the same documentation style — including its limitations. When an experiment is abandoned, it comes off this page rather than lingering as a phantom promise. Suggestions and bug reports feed this list directly: write to support.passport.room@gmail.com or use the contact page.

Try the current AI features free