What spends an AI size-chart read, and what costs nothing

A read is one call, not one photo and not one item

A size chart on a listing is usually a picture. It is a table of numbers, photographed or screenshotted, that no browser can search and no spreadsheet can import. An AI chart read is the one job that turns that picture back into numbers: the model looks at the frames, finds the rows, and returns them as text Credenza can measure you against.

The unit is the call, not the image. One card, one request, one read — however many frames rode along with it. That matters because the free plan allows 2 chart reads a day and Pro allows 15, and the natural worry is that a chart split across three photos eats three of them. It does not. Three frames go up together and come back as one answer.

The unit is also not the item. Credenza reuses a chart it has already read, so the second, fifth and tenth item from the same seller are all free. More on that below, because it is the reason the daily numbers are larger than they look.

Three things happen before any read is spent

Credenza does not send a request the moment you open a card. It works down a ladder, and every rung above the last one costs nothing.

• The text on your device. Many sellers type the chart into the description instead of photographing it. Credenza parses that text locally — no request leaves the device, so no read is spent, and it works with no signal at all. It reads labelled lines and simple tables, and accepts measurements between 20 cm and 250 cm so a price or a year cannot be mistaken for a chest.
• A chart you have already read. Before sending anything, Credenza looks along your own shelf for another item by the same seller whose chart was read rather than guessed. If it finds one, it reuses those numbers. This costs zero reads.
• Your own measurements. The size suggestion is computed on your device from the chart and your saved body profile. Comparing your chest against a chart is arithmetic, not AI, and it never costs anything — storing your measurements explains what it uses.

Only when the text has nothing usable, and the shelf has nothing to reuse, does Credenza offer to read the photos.

Up to 3 photos, or up to 10 album images, in one request

There are two ways a chart reaches the reader, and both count as one read.

Photos you attach. You can add up to 3 chart images to a card — a camera shot, a screenshot, or a saved image. Three is deliberate: it covers a chart split across two frames plus a retake. Each is compressed on your device before it is sent, so a large camera photo does not become a large upload.

Album images Credenza already has. If the listing was imported with a gallery, Credenza can scan up to 10 of those images for a chart without you photographing anything. It fetches them from the seller's own image host, not from a copy — turning an album into a list covers how those images arrive.

Either way it is one call. The read has 25 seconds on the server and 30 on your device before it gives up, which is generous for a photograph of a table and short enough that a stuck request does not hold the card open.

Your shelf is the cache, so a haul from one seller costs one read

This is the part worth understanding, because it changes what the daily number means in practice.

When a chart is read, Credenza tags it with the seller it came from and stores it on the card. Open a different card from that same seller and Credenza walks your shelf, finds the tagged chart, and uses it. No request is made. If several charts match, the newest wins, and a chart that was actually read always beats one that was guessed from text.

A ten-item haul from one Weidian seller therefore spends one read, not ten. A haul spread across four sellers spends four. Two a day on the free plan is not two garments — it is two sellers you have never measured before.

The cache lives on your device with the rest of the shelf. Nothing is stored on a server, and clearing the shelf clears it.

A miss counts. A failure does not.

Two outcomes look similar and are charged differently, so it is worth being plain about both.

A miss counts. If the model looks at your photos and answers "no size chart here", that is a read. The work was done and the answer is real — "this listing has no chart" is useful, and it cost the same as finding one. Photographing a blurry tag and hoping will spend reads.

A failure does not count. If the request times out, if the server returns an error, or if none of the album images can be fetched, nothing is charged. Not on the server and not on your device. You can try again immediately. This is a rule we had to fix rather than one we got right first: an earlier build counted on the device the moment a reply arrived, before checking whether the reply was an error, so two timeouts could spend a free user's whole day. They no longer can.

2 a day free, 15 a day on Pro, reset at midnight UTC

The counter is per day and it is the same counter on both plans, only with a different ceiling. The day boundary is midnight UTC, not your local midnight, so the reset is at a fixed hour wherever you are.

Nothing carries over. An unused read on Tuesday does not make three available on Wednesday. There is no bundle to buy and no top-up: the only way to have more in one day is Pro, and Pro's 15 is a ceiling nobody organizing their own hauls is likely to meet.

When you reach the free limit, the card does not break. It keeps whatever the local text parser found, keeps every chart already on your shelf, and keeps giving size suggestions from them. Only the new cloud read is unavailable, and only until the reset. Pricing lists this row next to the others.

How to make two a day be enough

• Import the listing before you photograph. The local parser only sees text Credenza has. Import first and it may find the chart with no read at all.
• Do the sellers you do not know first. The read is per seller in effect, so spend the day's reads on new sellers and let the cache cover the rest of the haul.
• Photograph the whole table. One clear frame of the entire chart beats three tight crops. The reader keeps original labels, including Chinese ones such as 胸围 and 衣长, so you do not need to translate before sending.
• Save your measurements once. A chart with nobody to compare it to is just a table. The suggestion is free and instant once the profile exists.

What the numbers are for

A chart on its own does not decide anything. Once Credenza has the rows, it compares them against your saved measurements and suggests a size for that garment, with the fit preference for that category applied — a jacket you want roomy and a tee you want close are measured against the same body with different targets.

The numbers stay on the card, so the decision is still there weeks later when the parcel arrives and you are checking QC photos against what you expected. Reading a Weidian size chart covers what the columns mean, and tracking QC photos covers the check at the other end.