
QR codes in AI product images can look convincing without doing the job you intended. A coffee pouch may retain a neat square pattern while its information changes. An approved campaign code may work in your large design file but become unusable in the version a shopper actually sees.
Before improving the prompt, separate three questions: Does a code belong in this placement? Does the final image decode correctly? Does the decoded information lead to the intended destination? A pleasing image, a successful decode, and an appropriate customer journey are three different approvals.
This guide gives you a source record, a decision tree, and a final-file checklist. For the wider creative process, see our AI product photography overview; references and generation should still be treated as ways to produce candidates, not guarantees that packaging remains unchanged.
The cover is an AI-generated editorial illustration of a fictional coffee pouch, an inactive phone, a blank card, and a magnifying glass. It contains no functional QR demonstration. All scenarios below are hypothetical: we did not run product-generation experiments or scanning tests for this article. Official sources were checked on October 28, 2026.
Start by naming the code's role. Is it real print on the package, a separate campaign call to action, or merely a decorative pattern the model invented? Do not approve the third as either of the first two.
Use this decision sequence before editing:
Keep separate exports for separate placements. A campaign card and a catalog main image can share a visual theme without sharing every graphic. Approving one file for social use does not approve it for a shopping feed.
The most useful handoff is not “keep the QR sharp.” It is a small record connecting the asset to its purpose. Ask the product or campaign owner to confirm:
Keep the code type clear. A retail UPC barcode and a campaign QR containing a web address are not interchangeable assets. GS1 US's barcode artwork guidance emphasizes appropriate artwork and printing for actual product barcodes. Its packaging instructions are not a universal specification for a QR shown on a phone screen.
For an unfamiliar code, inspect the decoded value before following it. If the value differs from the owner's approved record, stop and resolve the discrepancy; opening an unexpected destination is not a necessary quality-control step. Record the mismatch instead.
The packaging label accuracy guide covers the surrounding print and product evidence. Here, the extra requirement is a traceable relationship between source asset, encoded information, and intended use.
For a separate campaign graphic, a conservative workflow is to finish the creative scene first, then place the approved code in a suitable external image editor. Keep its source file available and its layer separate. Do not ask a model to redraw it as a finishing touch.
You can explore a scene or background in Oxava, then move the approved candidate into the external layout workflow. QR encoding, decoded-value inspection, and deterministic compositing are separate tasks; this article does not claim Oxava performs or certifies them.
For print already on a product, preserve the actual package rather than pasting a convenient code onto a newly invented bottle or pouch. If the photograph cannot support the required view, obtain a better source or evaluate verified artwork and geometry. A correct code on an inaccurate package is still an inaccurate product image.
In the external editor, keep the code's clear surroundings intact. DENSO WAVE's QR code area guidance specifies a clear margin of four modules on every side of a standard QR symbol. A module is one cell of the symbol: “four modules” does not mean four pixels in every export. Do not crop the margin to make the design look tighter.
Keep logos, decorative frames, and product shadows out of that clear area. Avoid stretching the code into a rectangle or treating it as an artistic texture. These are conservative production choices, not a claim that one layout recipe guarantees a scan.
Do not use error correction as permission to cover an arbitrary percentage of the image. DENSO WAVE's error-correction explanation describes redundancy and a data-size trade-off; its restoration percentages concern codewords, not a freely removable portion of the picture. A logo placed over an important region is not validated by a percentage remembered from a tutorial.
If you restyle, inpaint, or generatively upscale the assembled image afterward, treat the code as unapproved again. The safest checkpoint is after the last operation that can change its visible pattern, not before the most attractive enhancement.
“The QR does not work” can describe several different failures. Use the first failing stage to decide who should act next.
Compare the final file with the approved source. Look for changed cells, a cropped margin, blur, a tiny display, strong perspective, glare, or an obstruction. Change one suspected cause at a time and retain the previous version for comparison.
If the source asset itself is uncertain, return to the owner or encoder. Do not spend rounds sharpening an invented pattern. If the source is sound but the composition makes it too small, revise the layout rather than promising that higher export resolution alone will solve the problem.
This is a source or mapping problem, not a sharpness problem. Compare the raw value with the record. Check whether the wrong campaign, package revision, or product asset was used. A functioning code for the wrong lamp manual is not a partial success.
Replace the incorrect asset from the approved source, then repeat the final-file checks. Do not edit individual squares by eye to make one code resemble another.
For a verified URL, distinguish the encoded address from what happens after navigation. Have the responsible owner check the expected page, any authorized redirects, access requirements, and availability. A page that requires a customer login may be intentional or may defeat the campaign's purpose; the creative team should not guess.
A new image export will not repair an unavailable landing page. Conversely, a working homepage does not establish that the specific instruction page is correct. Record the actual destination checked and the time, not simply “website works.”
Compare the working image with the asset retrieved from the intended delivery surface. Check the crop and display size, not just the filename. Resize or compression may be worth investigating, but establish the difference before naming the cause.
The product image export checklist covers the broader source-to-storefront chain. For the QR, repeat the decode and value comparison on the delivered derivative. A screenshot from the design canvas is not evidence about that derivative.
The following are planning examples, not successful scans or measured production outcomes.
A roaster supplies a photograph of its current pouch with a small code on the side. The brief asks for a cleaner catalog scene, not a new scanning campaign. The owner confirms the source package and code record.
Decision: preserve the photographed package and review it against the source. Do not enlarge the code by replacing physical print with a different badge. Do not add a second promotional QR to the Merchant main image. If customers need brewing instructions, provide an appropriate separate link or eligible supporting placement rather than forcing the main image to do every job.
A lamp brand wants an approved campaign card that directs viewers to the correct assembly guide. The lamp photograph is accurate, but the first layout contains an AI-generated square pattern.
Decision: reject that pattern as an unverified code. Obtain an encoded asset from the confirmed instruction URL, keep it on a separate card outside the product, and check that the campaign placement permits it. Compare the decoded value and final destination with the specific lamp revision. For a phone-first audience, add a direct link where supported instead of depending entirely on scanning.
A skincare concept preserves the bottle silhouette but changes a printed code and part of the label. The grid looks tidy when viewed at normal size.
Decision: hold the image. Return to the real photograph or approved artwork; do not infer that tidy cells preserve information. The code review cannot excuse altered ingredients, quantity, or packaging geometry. If the required angle lacks sufficient evidence, revise the brief or obtain a new source before continuing.
Use a short record for each output, not one blanket “QR approved” label across a campaign:
Repeat relevant checks after a changed crop, asset replacement, campaign destination change, or new delivery format. Keep separate outcomes: a successful scan does not prove product authenticity, platform eligibility, accessibility, conversion, or an eventual purchase.
For many products, attach this record to the SKU image quality-control workflow. That makes it easier to find which image and revision were actually reviewed without implying every future derivative is covered.
If you later use the image in video, a still approval does not approve the clip. Check the final encoded video at the intended viewing size and review whether the code remains stable and usable for its purpose. This guide sets no universal hold duration and reports no video scanning test. Retain a suitable alternative link.
The useful question is not “Does this look like a QR code?” It is “Is this the approved asset, in an appropriate placement, returning the expected data from the version customers receive?” Keep those decisions visible and a failed check has a clear next step.
Start with a verified product source, develop the creative scene in Oxava, and handle the code in the appropriate external encoding and editing workflow. Then approve the actual delivery. An attractive scene is a creative result; a correct, usable customer route needs its own evidence.
A convincing square pattern is not evidence that the image contains your intended encoded data. Compare it with the approved source asset, then check the delivered file rather than only the working master. Cropping, blur, scale, or a changed pattern need different fixes.
Do not rely on a generated repair to recover unknown data. Obtain the approved code asset and its expected decoded value from the owner, or recreate the code from that verified value in a suitable encoder. Check the result independently before using it.
Google's image_link guidance explicitly lists QR codes overlaid on the photo among prohibited promotional elements. A code physically printed on the actual product is a different situation, but that does not automatically make the complete image eligible. Review the current requirements for the exact placement.
There is no single pixel size established by this guide for every code, platform, and viewing condition. Preserve the required clear margin and test the actual delivered asset at its intended display size. Do not treat a large desktop preview as proof of a usable mobile version.
No. It shows that a particular scanner decoded something under the conditions checked. Compare the decoded value and intended destination separately, then review the product's label, shape, variant, and other visible facts against the approved source.
Be the first to hear about new techniques, model updates and ideas on AI generation.