How We Test Photos, Sizes and Stencils

This page is the working method behind the photo check, the size guidance and the stencil export. It is written so you can judge how much weight to put on each claim — including the parts we have not measured yet.

What the photo check measures

The check runs on a 256 px copy of your image, in your browser. It does not upload the file. Four numbers come out of it:

Resolution

The longest edge of the file. Bands: 2000 px or more is plenty, 1000–1999 is workable, 600–999 is small, below 600 is too small for a face or lettering. Resolution is necessary but not sufficient: a sharp 4000 px photo beats a blurry one, and a blurry 4000 px photo still scores well on this check.

Contrast

The standard deviation of luminance on the 256 px copy. Bands: 40 or more is good, 28–39 is flat, below 28 is very flat. Flat light is common in phone photos of pets and in old family portraits; it makes the outline guess instead of trace.

Background clutter

The density of strong edges in the outer ring of the frame compared with the middle. A quiet background scores 0.60 or lower; 0.60–0.85 has some texture; above 0.85 is busy. Busy backgrounds are the most common reason a pet photo converts badly, because the edge detector reads the sofa, the rug or the garden as part of the animal.

Subject in frame

The area of the box that holds most of the strong edges, or a box you drag yourself. 45–85% is a good crop; 25–44% is small; below 25% is too little; above 85% is tight. At 6 cm, a face that fills a third of the frame has very few pixels left for the eyes.

What it cannot tell you

• Whether it will look like the subject. That is the conversion’s job, not the file’s. A high score does not guarantee likeness.
• How much detail survives at your size. That depends on the style, the line weight and the placement. The size page gives practical ranges; a measured atlas is in progress.
• How ink will read on your skin. Skin tone, texture, sun exposure and the artist’s technique all change the result. The try-on swatches are screen approximations, not a prediction.
• Sharpness. A 4000 px blurry photo scores at the top on resolution. Zoom to 100%: if the eyes or the letter edges are mush, they will not get sharper downstream.

How the bands were set

The bands are hand-set heuristics, not fitted to a labelled dataset. We sanity-checked the arithmetic on the site’s own sample photos and on synthetic edge cases — flat light, a noisy background, a small subject — which is enough to catch nonsense and not enough to call it accuracy. We say this on the page because a check that pretends to be a measurement is worse than no check.

The next version is a published benchmark: 300 real pet and portrait photos, rated blind by three tattoo artists at 6, 10 and 15 cm, compared with the four numbers above. The output will include the agreement rate, the false positives and the cases where the check was wrong. Until that exists, use the check to catch obvious problems, not to certify a photo.

How we test the stencil

The stencil is exported as a 300 DPI PDF at the width you choose, mirrored for transfer. The export is built to meet three conditions before we call it usable:

1. Physical size. The printed width matches the centimetres selected in the tool, not the screen size and not “fit to page”.
2. Line separation. The smallest gap in the design is at least twice the line width at final size, so it does not close when the skin heals.
3. Orientation. The design is mirrored for transfer and the text, if any, is readable after transfer — not before.

We have not yet published a physical print test at multiple sizes; that is part of the next benchmark. We do not claim that a stencil is a finished tattoo design. Most artists redraw or adapt the file to the body, the skin and the way the design will age. The seven handoff checks are the list we use before a file goes to an artist.

Corrections and versioning

If a page is wrong, we want to fix it and say so. Email support@tattoofromphoto.ink with the URL and the correction. When a method changes, we update this page and note the date. The current version is 2026-11-01.

What we publish next

• Photo readiness benchmark — 300 photos, three artist raters, agreement and error rates.
• Size and fidelity atlas — 24 photos at 5, 8, 12 and 18 cm, printed and photographed at 1:1.
• Healed case library — real clients, with consent, at 0, 3, 6 and 12 months.

These are listed as work in progress, not results. When they are published, this page will link to the data and the method, and the claims on the rest of the site will be updated to match.