Apple Now Adds SynthID to iPhone AI Images. Who Can Check It, and What Does a Result Mean?
Since iOS 27.2, most images that Apple Intelligence tools create or edit carry a hidden SynthID watermark, according to Apple's iPhone User Guide. The iPhone itself cannot read that watermark. To check it, you upload the file to an outside tool, and each tool answers a narrower question than "is this AI?"
So if someone sends you an iPhone image and you want to know whether AI touched it, the phone will not tell you. This guide walks through where you can check, how to read the answer, and what the answer still leaves open. It is written for anyone who receives images from outside, from platform and newsroom teams to rights holders.
What changed on the iPhone?
Apple's User Guide says that on iOS 27.2 and later, most images generated or edited with Apple Intelligence tools get a hidden SynthID watermark, which can be checked with the SynthID Detector. MacRumors reported that images made with Image Playground and some photos edited with Apple Intelligence now include the watermark.
Note the word "most." In MacRumors' test, photos changed with Extend and Reframe carried the watermark, while Clean Up depended on the size of the edit, and small edits had no watermark added. MacRumors also noted that there is no option on the iPhone to detect SynthID watermarks.
Where can you check an image for SynthID?
Since the phone cannot do it, you need an outside tool. Google opened the SynthID Detector to everyone on October 7, 2026, available globally in English. In the announcement, updated on October 29, Google says the Detector checks for AI from Google and its partners, including OpenAI, NVIDIA, Kakao and Apple. Gemini's own verification is narrower: its help page says Gemini can currently only recognize content created by Google AI tools.
Where | Whose watermarks it reads | Limits reported |
|---|---|---|
SynthID Detector (synthid.com) | Google and partners, including OpenAI, NVIDIA, Kakao and Apple (per Google, Oct 29 update) | Sign-in required; about 10 checks a day per user (Ars Technica); English |
Gemini app | Google AI only; also reads Content Credentials on web and Android | One file at a time; about 10 image checks per 24 hours |
iPhone Photos app | None | No built-in check (MacRumors) |
Google engineers told Ars Technica the daily limit is there to keep people from using the checks to build SynthID removal tools. The limit makes sense for a public tool. It also means the Detector suits spot checks, not screening every image a team receives.
What does each result actually tell you?
Once you run a check, the answer needs careful reading. A SynthID check looks for one specific watermark. The Detector's own page says it is not a general AI detector. So each result answers a different question than the one people usually ask.
Result | What you can say | What you cannot say |
|---|---|---|
Watermark detected | All or part of the file was likely made or edited with a SynthID-enabled tool | That the whole picture, or what it shows, is fabricated |
Not detected | No confident SynthID signal was found | That the image is a real photo; it may come from a tool without SynthID, or heavy edits may have weakened the signal |
Unsure | The file did not give enough to decide, for example a very small edit | Either answer |
The Detector's FAQ also says that in very rare cases SynthID can falsely trigger on content without a watermark. A positive result is strong evidence, not a verdict.
How should a team handle images from iPhones?
In our view, the practical question for a receiving team is at which step someone will read the watermark and what they record. Three habits help.
Check the original file. Ask contributors for the file they exported, not a screenshot or a messenger copy, and log where it came from.
Read two signals, not one. A watermark is hidden in the picture itself. Content Credentials (C2PA) are a signed label stored with the file that says where it came from and how it was edited. Gemini's help page treats them as separate checks, and either can be missing.
Write down the result as it was given. "SynthID not detected on [date] via [tool]" is accurate. "Verified real" is not.
For images you publish yourself, the order flips: you decide what goes into the file before it leaves. Image Defence, part of MuseBlossom's Contents Defence, applies an invisible watermark and issues and verifies C2PA Content Credentials for your own images. Contents Defence is on the C2PA Conforming Products List for image, audio and video; a listing shows conformance to the specification, not how well a product performs.
What can't a SynthID check tell you?
It cannot tell you whether an image came from an AI tool that does not use SynthID, and it cannot tell you which part of an image was edited; the public Detector returns only whether SynthID was found. It also says nothing about whether a photo's content is true. Those questions need other evidence: the original file, the sender, and the context it arrived in.
The short version
An iPhone now marks most of the AI images it makes, but it cannot read those marks back. When an image matters, get the original file, run it through the SynthID Detector, and write down exactly what the tool said. Treat the answer as one useful clue, and keep looking at where the file came from and who sent it.
FAQ
Does every AI edit on an iPhone get a SynthID watermark?
No. Apple's guide says most images, not all. In MacRumors' test, small Clean Up edits had no watermark added.
Can the Gemini app check an image made on an iPhone?
Gemini's help page says it can currently only recognize content created by Google AI tools. Use the SynthID Detector for partner watermarks.
Is the SynthID Detector free?
Google's announcement opens it to everyone, and reports describe a daily quota of about 10 checks after sign-in. We did not find a price in Google's announcement.
If no watermark is found, is the photo real?
Not necessarily. The Detector's FAQ says a "not detected" result does not totally rule out AI generation or manipulation.
Adding a watermark and Content Credentials to the images you publish? Ask us about Image Defence.