Pull 15+ data points from any label

A reseller in our beta cohort once spent forty-five minutes on a single Levi's 501 listing because the inside-collar tag had three sizing systems printed on it. W30 L32. The Asian "Size L." A faded EU 42 in a font the camera kept reading as 47. Every minute of that was data already on the label. The label held the answer. The keyboard slowed her down.

That gap is the whole reason the Label Scanner exists. Point the camera at any garment label, hold steady for a second, and Size AI extracts more than 15 structured data points before you can put the phone down. Brand. Multi-format sizing. Fabric percentages. Fit type. Care symbols. Manufacturing details. The label always told the buyer that information. Now it tells the listing too.

What follows is what the Label Scanner pulls, how to scan a label cleanly, and a handful of edge cases worth knowing about.

What the scanner actually does

Point at any label and the on-device OCR engine reads the text. A second model classifies what each piece of text means: this string is a brand, this is a size in EU format, this is a fabric percentage, this row of pictograms is a wash code. The structured output saves to the garment record automatically.

A single scan typically lifts 15+ data points off a single tag. OCR runs at 95%+ accuracy across faded labels, vintage tags, and the kind of warped print that shows up after a few cycles in the dryer. It runs on-device, which is why scanning works in a thrift-store basement with no signal.

What gets extracted

Scanning a clean label produces:

• Brand and model. Full names plus model where available, like "Levi's 501" or "Patagonia Better Sweater"
• Multi-format sizing. W30L32, Size L, EU 42, UK 10, all read off the same tag where present
• Fabric composition. Each fiber and its exact percentage, like Cotton 84%, Polyester 15%, Spandex 1%
• 5-level stretch analysis. Minimal, Light, Medium, High, or Maximum, derived from fabric content and construction
• Fit classification. 17 fit types from Skinny to Relaxed, cross-referenced against the actual measured garment
• Manufacturing details. Country of origin, production season, factory codes, RN numbers
• Care instructions and symbols. Wash temperature, drying method, ironing temperature, plus the pictograms most buyers don't bother to decode

Care instructions matter more than they sound. They're the second-most-asked question in marketplace messages after sizing. Including them in the listing description preempts the message thread.

How to scan a label

The flow lives in the Results screen after a measurement capture.

1. Tap "Scan Label." The camera view opens with a focus frame in the center.
2. Fill 50–75% of the frame with the label. Hold the phone parallel to the tag, not at an angle. Daylight is best. Avoid glare on glossy tags.
3. Hold steady for one to two seconds. The scan completes automatically when the OCR is confident. No shutter button.
4. Review the extracted fields. The app shows every detected data point with a confidence score. Edit anything that looks off, then save.
5. Scan additional labels for the same garment. Some pieces have a brand label inside the collar, a fabric content label on the side seam, and a care label in the wash bag. Scan each one. The app merges the data into a single garment record.

Scanning one clean label takes one to two seconds. Multiple tags per garment add up to maybe ten seconds total. End to end, the saving versus typing data into a listing template runs about thirty to sixty seconds per garment.

Tips that actually move the needle

• Flat label first. The OCR handles light bending, but a tag folded at the seam often misreads sizing. Pull the label flat against the table or against the inside of the collar before the scan.
• Multiple tags merge cleanly. Scan the brand label, then the fabric tag, then the care tag. The app reconciles them into one garment record without overwriting fields that already have higher-confidence data.
• 8 languages on the OCR. English, Spanish, French, German, Italian, Portuguese, Japanese, and Chinese all read correctly out of the box. The structured output is in your app language regardless of the language printed on the tag.
• No internet required. The OCR pipeline runs on-device. Scanning works in a thrift store, a warehouse, a flea market, anywhere.
• Manual fallback for missing labels. Cut tags and completely missing labels can't be scanned. The Garment Detail screen has a manual entry fallback for every field the scanner would have populated.

Common questions

Does the Label Scanner work on faded or vintage labels?

Yes. The OCR is tuned for the wear typical on resale inventory: faded screen prints, raised heat-transfer text that has cracked, woven labels with thread pulls. Accuracy holds at 95%+ across that condition range. Heavily damaged labels with missing letters fall back to a confidence score under 80%, which the app surfaces so you know to check the field manually.

What about international or non-English labels?

8 languages are supported on the OCR pipeline directly: English, Spanish, French, German, Italian, Portuguese, Japanese, and Chinese. Labels printed in those languages read correctly without any setting change. Other languages still extract sizing and fabric percentages, since those rely on numerals and pictograms more than text recognition.

What if the garment has no label at all?

The Garment Detail screen has manual entry fields for every attribute the scanner would have populated. Brand, sizing, fabric composition, fit type, care instructions are all editable directly. The scanner is the fast path; manual entry is the always-available fallback.

How accurate is the OCR?

95%+ across our internal validation set, which includes faded labels, vintage tags, multi-language labels, and labels with mixed text and pictograms. Brand and model recognition trends higher (98%+); care symbol interpretation trends slightly lower (94%) because some legacy symbols have multiple regional interpretations.

Does it work without an internet connection?

Yes. Both the OCR and the structured-output classifier run on-device. The scanner works anywhere your iPhone runs, including basements, warehouses, and flea markets with no signal. Cloud-side features like AI Photo Studio need a connection; the Label Scanner does not.

Related

• Capture. The measurement flow that runs first in the same Results screen.
• AI Descriptions. The label data flows directly into the description generator, which uses fabric, fit, and care fields to write listing copy that reads as informed.
• Building a marketplace, retail tech stack, or fit-data layer? Talk to our team about API and SDK access.