Latest updates and improvements to the platform
Packshots generated before July 2026 are no longer refused the label as photos you uploaded yourself.
After you replace a packshot's image, downloading it with the AI label now gives you the new image, not the one it replaced.
The visible AI label can now read AI MODEL, word for word, in every language. The wording you use today is unchanged.
When the visible AI label cannot be added to a photo, you are told right away instead of after a three-minute wait.
When your AI label describes the model, packshots and photos without a model now get a wording that describes no one. You choose it.
A labelled download of a photo generated before mid-July 2026 no longer ends without the label.
A file you add with Upload Own or Use Original is now converted and signed before it can be downloaded. While that happens its tile says so, and if it fails you can try again or remove it, with your file kept.
Ordering a packshot moments after uploading a photo used to fail at once, because the photo was still being prepared. The order now waits and starts on its own. On a tile that did fail, the button that archives it no longer sits on top of the one that retries.
The LAVEL case study has been rewritten on the client's latest figures. Its results now come in two windows — a clean comparison of the first quarters and January 1 to July 31 year over year — alongside the Meta Ads and Google Merchant Center metrics for the same period.
A product whose name ran past 200 characters had every generation refused, with the credits returned but no pictures produced. Names are now limited where you type them, and a name already stored no longer stops anything.
An invitation sent to an address containing capital letters could not be accepted — the link led to a bare "page not found". The address is now matched regardless of letter case, and a link that genuinely no longer works says so.
After signing in, every member of a team finds a seven-part package on labelling photos and videos generated with AI on the team's home page.
When the service that generates your images applies a short-term limit, the platform now waits and retries, instead of treating the refusal as a problem with the provider account and stopping on the first attempt.
When the service that generates your images runs out of capacity, the platform now says so and suggests a realistic wait, instead of describing it as a brief hiccup that will pass in a moment.
When our image provider refused one of your photos, the generation kept retrying and then failed with a message that read like a temporary fault. It now stops at the first attempt and tells you which photo to replace.
Uploading a photo that arrived carrying a signed AI marking could fail outright, because the marking was being lost when we converted the file. Those uploads now go through, and the marking travels into the packshot as the source ours sits on top of.
Retrying more than one failed image from the same order used to lock part of your balance permanently and make the second retry fail. Both are fixed, and every credit held that way has been returned.
The comparison slider, the marketplace filters and the demo wizard are operable from the keyboard, every control names itself in all three languages, and autoplaying video and the scrolling rows can be stopped.
A file's statement about how it was made now counts only when it describes the picture itself. A statement about an edit — a retouch, an upscale — no longer makes the visible AI label available for your own photograph.
A requested address change confirmed nothing: the link was used up and the address stayed as it was. The change now goes through on a single click, the page names the address your account moved to, and the previous address is told about it.
An uploaded photo could be offered the visible AI label because the file itself said it was AI-generated. That claim is now verified first, and one that cannot be verified no longer makes the label available.
The action that used to delete an account now takes it out of service: it stops working, any subscription stops renewing, and the records stay. Permanent erasure of your data is a separate request, carried out after we verify it.
On narrower windows the preview covered part of the form, putting some settings out of reach. The two contrast readings now sit with the colour pickers, and the button that corrects a poor pair appears where the warning is — it had never appeared at all.
The signed record travelling with every export was rebuilt into the shape the C2PA conformance programme requires. It now names us as its author, states the specification version it follows, and says plainly whether it lists every step we performed.
Several download buttons handed over a file without ever asking whether you wanted the visible AI label on it. Every download control now puts the choice in front of you first, and the plain file is still one click away.
When none of the images in a download could carry the visible AI label, the message said "Labelled 0 of N" and listed the filenames. It now names the reason instead.
If your browser remembered that you last signed in with a password, the sign-in page threw away the form it had just been sent and built a new one. The form now arrives in one piece and is ready to type into sooner.
A link to a Polish, Ukrainian or English documentation page used to open in whatever language had been stored from an earlier visit. Every documentation address now opens the version it names, and the language selector moves you between the versions of the page you are reading.
A packshot that could not start because the product photo was not ready reported an error on our side. It now names the actual reason, waits for you instead of repeating itself, and reports the same reason to shop integrations.
A note longer than 2000 characters used to make a photo-shoot submission fail with an unexplained error. It is now refused straight away, with a message naming the field and the limit.
The public label tool drew nothing in Safari — it reported that the badge typeface had not loaded and produced no preview and no download. It now confirms the typeface immediately before every badge it draws.
Wordings that name or describe the model used the feminine form for everyone. Newly generated model cards now record the model's gender, and exports use it. The public label tool asks the same question.
The label settings screen now hands you the badge on its own, as a PNG on a transparent background, at the size it has on a full-size photo. Useful when your shop puts the badge on photos itself.
The feedback form now has a category for reporting that a generated person resembles a real individual, or that content breaks the law. No email needed, and you are told when to expect an answer.
A generation that did not succeed on the first attempt now explains itself on the tile instead of showing a bare spinner, and a brief the model refuses fails at once with its credits returned.
Some files state how they were made, inside the file itself. We now read that when you upload, and stop describing a file as your own photograph when the file says otherwise.
Ordering a shoot of the front of a product could return the back. The shoot now follows the photo you selected, and the product card lets you choose it.
A product photo that already carries credentials from your editing software keeps them when you promote it with Use Original. Ours are added alongside, not on top.
A new public page asks at most seven questions about one photo or video and tells you whether it needs a visible AI label, with the legal basis for the answer. Your answers stay in your browser.
A product photo you promote to a packshot is now prepared and signed the same way as everything else you download here. Your original photo is left exactly as it was.
After checking an image you can now ask for a signed report of the result — a file to attach to a complaint or give to an auditor, which anyone can verify without us. The image itself is still never sent.
Styles and models made against our earlier standard have been withdrawn from the marketplace, several of them superseded by a newer version of the same idea. Anything already on your account stays there.
A new guide explains when a photograph needs an AI disclosure, gives four ready wordings to copy, says where to put them on each sales channel, and lists what you are not obliged to do.
Anyone can now open /tools/ai-label, drop in a product photo and see the label on it — no account, no charge. The photo is never uploaded.
A new free tool writes an origin declaration into the metadata of a JPEG you supply. The declaration is yours, the file never leaves your browser, and nothing about it is kept.
The result at /tools/verify-image now carries a short section of next steps, chosen by what the check found and by what each destination will actually accept.
The label can now say what was *not* generated, not only what was — including naming the model while stating the product is real. The wordings you already use are unchanged.
A new generation of styles is live, each with its own page, gallery and guidance on what it is best for. They are free to add to your account.
New accounts start with a curated set of six styles recorded against them, and every style on your account is now one you can remove. Styles that used to appear on everyone's account without being chosen no longer do.
Six list endpoints on the external API now accept a size, and every list says whether it was truncated instead of looking complete.