QR code to verify an AI product photo: scan to see the real item

The QR code opens that image's check page, with no sign-up or login. It shows the real product photo beside the image from the post, and tapping zooms in on fabric, colour and lettering. Extra real photos the shop added, such as a fabric close-up or the neck label, appear underneath.

• A badge reading "Shop-confirmed · compare with the real item" with the shop name and the date the image was made

• What the shop confirms matches the real item: print, colour and lettering (for other products: shape, colour and label)

• What the AI made new, such as the model (not a real person), the pose, the background, the light and shadow, and how the fabric drapes on the body

• Wording that changes with the tool: Ghost Mannequin says there is no person in the image, and a Product Showcase hands-only pose says the hands and scene are AI

• A note that the real fit on a body may differ slightly from the image

In this pair, the zigzag-edged collar, the black placket with gold buttons and the two chest pockets match in both photos. But the two black-trimmed pockets at the hips in the real photo appear as black bands with gold buttons at the waist in the AI image. That is the kind of thing a shop should spot before confirming: if the difference matters to buyers, generate the image again instead of confirming it. A buyer can compare the same way after scanning.

None of this costs credits. The image with the QR is assembled on the shop's own device, and the original AI image in the library is not changed. The downloaded image carries the card shown in the sample above. Turn it on again for the same image and the old link and QR keep working.

• The top two bold green lines, "Matches the photo" and "Verifiable", are fixed on every image.

• The third line names the image type and changes with the tool: "AI model", "AI image" (a hands-only pose), "AI-assisted try-on" (Virtual Try-On with a person photo the shop uploaded) or "AI-staged" (Ghost Mannequin).

• The last line, "Scan to see the real item", tells the buyer what to do with the QR.

• The QR takes about 14% of the image's short side. In our 4 October 2026 test, the image shrunk to 40% of its original size still scanned.

• The QR uses the highest error-correction level (H) in case the image is recompressed when posted, and points to a link of the form plienroob.com/v/ followed by 8 random characters.

• The shop can pick any of 4 corners. The default is bottom right; the sample above uses bottom left so it doesn't cover the shoes.

• Anyone with the link can open it, but the page is kept out of Google and the link's code is 8 random characters that can't be guessed.

• The page shows no email, username or AI model name.

• The shop can switch it off any time from the box under the result or from the image library, and the old link and QR then say the shop has closed this image.

• An image that is switched on isn't deleted automatically by the data-retention period until the shop closes it itself.

• The shop can see how many times it has been opened.

It suits images posted on Instagram, Facebook, LINE and the shop's own website. The app notes under the button that Shopee, Lazada and TikTok Shop don't allow a QR code in product images. For those platforms, use the original image without a QR, and always check each platform's current rules yourself.

Plien Roob checks an uploaded photo in two layers. The first measures file quality (size, sharpness, compression). The second has an AI read what is in the photo before it is stored: whether it is safe, what product it shows and how it was shot, and then compares that with the tool you have open.

Every uploaded photo is put to an AI as three separate questions. The first is whether it is explicit (a woman's nipple or areola, or anyone's genitals: the file is deleted at once and nothing is saved; a man's bare chest is fine). The second is age: the product slot blocks only photos that look like a minor, while the model slot has its own stricter age check. The third applies only to product photos: what product it is and how it was shot. The three are kept separate on purpose, because in testing a combined question caught only 3 of 26 photos of children.

• Products it can read: 7 fashion kinds (clothing, footwear, bag, headwear or glasses, necklace, earrings, ring, bracelet or watch), 7 kinds Product-Sexy handles, the 15 held or displayed kinds of Product Showcase, a person photo, or other.

• Ways of shooting it can read: flat, on a mannequin, worn, screenshot, box or advert, and several items.

• The card shows three steps (product kind, content, tool) and says whether the photo is a "match", "has a better fit elsewhere" or a "mismatch" for the tool you have open.

• A "Go to …" button takes you to the better tool and sets the product category for you. You can always carry on with the tool you had; the card suggests, it never blocks.

• Yellow notes under the card, such as: Virtual Try-On works for clothing only, Product-Sexy takes only photos with no one wearing the item, crop the app buttons off a screenshot first, and crop a multi-item photo down to one item.

A photo of a person wearing the garment doesn't fit every tool. The one the system knows it will refuse is Product-Sexy, which needs a photo of the garment alone. If it finds a person, it says so before holding credits and offers to take the person out. Ghost Mannequin accepts flat-lay, hanger and worn photos, and gives a garment holding a worn shape on white with neither a person nor a mannequin.

If Virtual Try-On or Product to Model has a photo refused by the AI provider, the system offers Product-Sexy with the same photo, which builds the model from words you tap and sends no photo of a real person. And when something is missing at the moment you press generate (no model chosen, no image, not enough credits), the button takes you to where you can fix it instead of staying silently disabled.