iOS 27.2: Apple now stamps AI-edited photos with Google's invisible SynthID watermark

What's going on

Apple now labels images created or altered with its AI features with an invisible watermark. The change arrived with iOS 27.2, which Cult of Mac says was released on a Monday in early November. It also applies to iPadOS 27.2, macOS 27.2 and visionOS 27.2. Apple confirmed the feature in a support document dated November 5.

Apple didn't build its own system. Instead, it uses Google's SynthID. Apple first announced this at its WWDC26 developer conference in June, alongside a revamped Image Playground that can now also generate photorealistic images. You don't have to do anything: there's no prompt, no toggle and no visible mark in the image. The watermark is added automatically during editing.

Which features add the watermark

According to Apple, these Apple Intelligence tools apply the watermark:

So the watermark doesn't just end up in images most people would call "AI images." Ordinary camera photos get it too if you edit them with one of these tools. Both Apple and 9to5Mac explicitly point this out.

Conflicting reports on Clean Up

When it comes to Clean Up, the sources disagree. Deskmodder, citing Apple's support document, says a single use of the feature is enough to trigger the watermark. Cult of Mac and Notebookcheck, on the other hand, point to MacRumors' testing: Extend and Reframing reliably added the watermark, while with Clean Up it depended on how big the edit was. Small touch-ups stayed unmarked, larger ones were marked.

Notebookcheck considers that threshold sensible: removing a piece of lint from a sweater doesn't make something an AI image. But as Cult of Mac points out, there's a catch: a quick small retouch can slip through without a label. Exactly where Apple draws the line hasn't been made public.

How SynthID works

Google developed SynthID so AI-generated content can be identified after the fact. The watermark is embedded directly in the pixels and is invisible to humans. It isn't stored in the metadata, the extra information in an image file such as capture date or camera model. Metadata is often lost when you share an image through a messaging app or take a screenshot. SynthID instead tweaks the pixel pattern so subtly that you can't see it, but software can still find it. Google says this doesn't degrade image quality and survives cropping, filters and compression. Those are the vendor's claims. The sources don't cite any independent robustness tests.

In its support document, Apple stresses that the human eye can't perceive the watermark and that the photo's appearance stays unchanged.

As for older images, Deskmodder reports that photos from before the update won't be watermarked after the fact. An old image only gets marked if you edit it again with one of the supported tools. In other words, your photo library isn't being retroactively tagged.

Checking only works through Google

There's no way to tell on the iPhone itself whether an image carries a SynthID watermark. The Photos app has no built-in detection. Apple points you to synthid.com, where you upload an image and get the result. According to Notebookcheck, Google opened this SynthID Detector to everyone in early October. So if you want to check an image, you have to upload it to a Google service.

You also need to read the result correctly, as several sources point out:

• Watermark found only means one of the supported AI tools was used. It could be a fully generated image or a real photo with just one object removed.
• No watermark doesn't prove an image is real. The detector only recognizes watermarks from Google and its partners. According to Cult of Mac, images from other AI programs or photos manipulated the old-fashioned way slip through the net.

Our take

The most notable point, according to Notebookcheck, is that Apple is adopting a competitor's technology instead of building its own. That's pragmatic: a watermark is only useful if as many vendors as possible use the same method and you can check it in one place. With Apple on board, SynthID gets a lot closer to being a real industry standard, especially since hundreds of millions of iPhones and Macs will get the update.

The limits are still obvious. The watermark only shows what a participating vendor has marked. Anyone faking images with other tools leaves no watermark behind. The unclear Clean Up threshold and the lack of on-device checking further limit its everyday usefulness. For you, that means: if you polish photos with Apple's AI tools, be aware that they may afterwards be identifiable as AI-edited, even though you can't see anything. As Cult of Mac puts it, SynthID is a clue when an image looks too good to be true, but not proof of authenticity or forgery.