ID Photo Maker

Choose a photo

Result and adjustments

How it works

Your photo is levelled from the eye positions, then cropped to the standard you pick using the top of the hair and the point of the chin. Every pixel is then sorted into background, hair, skin and clothing, which is what makes the background swap possible. Choosing a suit measures your shoulders and neck as well, and the garment is drawn to those measurements. Use the framing sliders only if the automatic crop sits wrong, then save a single photo or a full print sheet at 300 dpi.

There are three groups of controls: size, background and clothing. Hair colour, expression and skin retouching are not available. An ID photo has to match the person first of all, and a tool that edits the face lets you build a photograph that no longer does.

The sizes and where they come from

These are the presets and the numbers behind them. “Head length” is the distance from the top of the head to the chin, and “space above” is from the top of the head to the top edge of the picture. Each value sits near the middle of the range the relevant authority allows.

The Japanese passport figures come from the Ministry of Foreign Affairs specification of 45 mm by 35 mm with a head length of 34 ± 2 mm and 4 ± 2 mm above the head; the midpoints are used here. The My Number card uses the same dimensions. The driving licence figures are the midpoints of a 2.0 to 2.4 cm head and 0.2 to 0.6 cm above it. The resume size is not fixed by any law: it follows the convention used by Japanese photo booths and studios, where the head fills 70 to 80 per cent of the height. The head size slider moves that proportion either way.

What it can and cannot do

Nothing here generates an image. The models decide what each pixel is and where your body is; the cropping, replacing and drawing is then done with ordinary arithmetic. One model returns 478 facial landmarks, one sorts pixels into six classes, and a third — loaded only when you ask for a suit — returns the position of your shoulders. All of them run inside the browser through WebAssembly.

Reliable

Cropping and levelling are accurate enough to leave alone. Eye positions are precise, and the crown and chin come out consistently from the hair and face masks. Background replacement recomputes the colour of the semi-transparent pixels along the hair, so the old background does not leave a fringe and you do not need a white wall to shoot against.

How the suit is fitted

What is composited is a photograph of a real suit. A product shot with the neck cut out of it — a ghost mannequin shot — is prepared in advance, and its neck opening is matched to your neck by scaling and rotating the image. The weave of the cloth, the way the collar falls and the sheen on the tie are all from that photograph, so the result is not a flat fill of colour.

Two cues do the fitting. The first is your neck: the width of the skin is measured row by row below the chin and the narrowest row is taken as the neck. Reading a single row just under the chin would measure the jaw instead, and the collar would come out as wide as the face. The second is the position of each shoulder returned by the pose model, which is used to rotate the suit to your shoulder tilt.

The height of the collar comes from the distance below your chin and the frame of the size you picked. The collar line of what you are already wearing — the row where skin stops — is only used when it sits higher than that, as a turtleneck does. Matching the collar to the neckline of a low-cut t-shirt would push the whole suit out of the frame on a tight size such as the 30×40 mm résumé photo.

The garment is laid behind you, with your face and hair kept in front, so not a single pixel of your face changes. The neck between the chin and the collar is built by continuing the skin tone from just under your jaw; without that step the collar of the t-shirt you were wearing ends up inside the collar of the suit. When the shoulders cannot be found — a photograph cropped tightly to the face, for instance — the width is estimated from the face instead and the fit is noticeably worse.

Not possible

It cannot restyle hair or rebuild an expression. Those need a model that repaints the photograph, and tested against real photographs the face comes out as a different person — which is exactly what an ID photo must not do. The suit works the same way for the same reason: rather than generating an image on the spot, one prepared photograph is fitted to your body. Everyone gets the same suit, and in exchange nothing at all is done to your face.

Skin smoothing, hair colour and expression adjustments were offered here until August 2026 and have all been removed: the list of controls had grown to the point where it was hard to know what to touch, and enough retouching produced a photograph that no longer matched the person. If you are already wearing a jacket, keeping the photo as shot is the most natural result of all — the cloth and collar are genuinely yours, and they fit your build exactly.

Using it for official documents

Japanese passport, My Number card and driving licence applications all state that retouched photographs are not accepted. A replaced background or composited clothing can be refused at the counter, or accepted and then sent back later.

Selecting one of those sizes therefore turns on official document mode, which performs the crop and the tilt correction and nothing else; the background and clothing controls disappear. You can switch the editing back on, but use that result for a resume, a staff card, a club roster or anywhere the recipient permits edited photographs.

The shot itself has requirements this page cannot fix: no hat, facing forward, plain background, taken within the last six months, eyes not covered by hair, no shadows. Check the instructions from whoever receives the photo. The measurements in the table above are drawn from published specifications, but the final authority is each institution's own documentation.

What happens to your image

The photo you choose stays inside your browser. Reading, face detection, cropping, editing and saving all happen on this page; the image is never sent to a server and never stored. Unlike the Base64 converter, this page logs nothing at all.

The model files themselves are fetched from outside. The library comes from the jsDelivr CDN and the trained models from Google's public storage, downloaded after the page opens — around 30 MB in total, cached by the browser afterwards. Those requests ask for files and carry none of your image data. The site-wide policy is in the privacy policy.

Printing

“Fill a sheet and save” tiles the photo across the paper you choose. Both paper orientations are tried and the one that fits more copies wins. On L-size paper that means 8 copies at 30×40 mm, 4 at 35×45 mm and 12 at 24×30 mm, with 1 mm between photos, a 2 mm margin around the edge, and light grey cut lines.

Printing one L-size photo at a Japanese convenience store therefore yields several ID photos for the price of one print. Choose the full size, with border setting rather than borderless: automatic enlargement throws the measurements off. The exported file carries a 300 dpi resolution tag, so software that reads it prints at the real size.

Related pages

If you are preparing for a job in Japan, the income threshold simulator shows where the tax and social insurance boundaries fall. Municipalities within a radius helps when the job means moving and you want candidates within a distance of the workplace. The rest of the site is listed under all tools. The Japanese version of this page is at 証明写真メーカー.

Frequently asked questions

Is my photo uploaded anywhere?

No. Reading the file, locating the face, cropping, editing and saving all happen inside your browser. The picture is never sent to this site and is never stored anywhere. The AI model files themselves are downloaded from jsDelivr and from Google storage after the page opens, but those requests only ask for the model files and carry none of your image data.

Which photo sizes are supported?

Four presets: 30 by 40 mm for Japanese resumes and job applications, 35 by 45 mm for a Japanese passport and My Number card, 24 by 30 mm for a Japanese driving licence, and 51 by 51 mm for a United States visa or ESTA photo. Any other size can be typed in as a width and a height in millimetres. Every size is exported at the equivalent of 300 dpi.

Can I use the result for a passport application?

Only the cropping and the tilt correction. Japanese passport, My Number and driving licence applications all reject retouched photographs, which includes a replaced background and composited clothing. Picking one of those sizes therefore switches on official document mode, which hides the background and clothing controls. You can turn the editing back on, but then use the result for a resume, a staff card or another purpose where the recipient allows edited photographs.

Why are there only three things to change?

Because the three things actually wrong with a snapshot used as an ID photo are the proportions, the room behind you and the clothes you happened to wear. This page used to offer hair colour, stray-hair trimming, mouth corners, eye opening, skin smoothing and brightness as well. The longer that list grew, the harder it became to know which control to touch, and it was possible to retouch a face until it no longer matched the person. An ID photo has to match you first of all, so everything that touched the face itself has been removed. What is left exists only to save you from taking the photograph again.

How realistic is the composited suit?

What is composited is a photograph of a real suit. One prepared image is scaled and rotated to your own neck width and shoulder tilt, so the weave of the cloth, the way the collar falls and the sheen on the tie are all from that photograph rather than flat colour. Your neck is measured from the shoulder positions returned by the pose model and from the width of the skin below the chin. The height of the collar comes from the distance below your chin and the frame of the size you picked; the collar line of what you are already wearing is only used when it sits higher, as a turtleneck does. The suit is laid behind you and your face and hair stay in front, so not a single pixel of your face changes. The neck between the chin and the collar is built by continuing the skin tone from just under your jaw. It works least well when your original clothing fills much of the frame, or when your body is strongly turned.

What if the suit does not fit properly?

Start with a photograph that includes your shoulders. If only the face is in frame there is nothing to measure, and the suit falls back to an estimate from the width of your face, which is much less accurate. Face the camera with both shoulders level; a strongly turned body puts one shoulder in front of the other and the outline no longer matches. A turtleneck or a hood hides the neck, which leaves less of it to measure. If it still looks wrong, lower the head size slider a little so more of the shoulders comes into the frame.

Which background colour should I choose?

White or light grey is the safe answer for a Japanese resume or job application, and white is never wrong when no colour is specified. Pale blue and sky blue are the traditional choice in Japanese photo studios and stop a white shirt from merging into the background. Deep blue separates light hair from the background but reads as showy to some recipients, so pick it deliberately. None of the options add a shadow, because every ID photo specification asks for a plain background with no shadow on it.

Why is the first run slow?

Because roughly 30 megabytes of AI models are downloaded for the face landmarks and for the hair, skin and clothing segmentation. Choosing to composite a suit adds about 6 megabytes more for the shoulder-detection model. Later runs read them from the browser cache and start almost immediately. On a slow connection the first load can take tens of seconds. Once the download finishes, the processing itself needs no network at all.

What if no face is detected?

Use a photo taken from the front with the face filling a reasonable part of the frame. Remove hats, masks and sunglasses, and keep the eyes clear of the fringe. Profiles, heads tilted far down, and crops out of group photos where the face is tiny will not be detected. Dim photos are also difficult, so retaking the shot near a window often fixes it.

The measurements are reference values taken from published specifications. Requirements differ between institutions, so check the instructions of whoever receives the photo before you apply. No liability is accepted if an edited photograph is refused.