This article may contain affiliate links. Learn more on our Partner Program page.
Anything created by Image Playground or edited by Apple Intelligence in the Photos app carries an invisible watermark. It originates from Google and is embedded directly in the pixels. However, it cannot be verified on the iPhone itself.
Apple announced the tagging feature in June with the next generation of Apple Intelligence. Images from Image Playground automatically contain an invisible SynthID watermark, as do photos edited with the AI tools that the Photos app received in iOS 27.
Key Facts at a Glance
- Images from Image Playground and photos edited with Apple Intelligence carry an invisible SynthID watermark.
- The method comes from Google and writes the marker into the pixels, not into the metadata.
- There is no verification on the iPhone; this can only be done via Google's SynthID Detector in the browser.
Pixels instead of metadata
From the very start of Image Playground with iOS 18, Apple recorded AI-generated images in the metadata, i.e., in the additional information of the image file. SynthID, on the other hand, is attached to the image content itself and, according to Google, should withstand common changes such as cropping, filters, and compression.
Back then, Image Playground exclusively produced cartoon images. Since iOS 27, it can also create photorealistic images, and the Photos app uses Extend, Spatial Reframing, and "Clean Up" to add image content that was never captured in that way. These features require Apple Intelligence, meaning at least an iPhone 15 Pro is needed.
This is how an image can be checked
The check is performed using Google's SynthID Detector, which identifies images, videos, and audio from providers that use the method. Registration is required, including via an Apple account, and the number of requests per day is limited. According to Google, uploaded files are deleted immediately after the result is available.
Evidence in one direction
In the EU, Article 50 of the AI Act has required since August 2, 2026, that providers of generative AI mark their outputs as artificially generated in a machine-readable manner. A watermark in the pixels fulfills this purpose significantly more robustly than an entry in the metadata.
I still believe the marker is only meaningful in one direction. If the detector finds the symbol, AI involvement is confirmed; if it's missing, that doesn't prove an image is unedited – Google itself points out that the detector only recognizes content from providers using SynthID.
Would you actually send a suspicious image through the detector, or is registering with Google for that too much hassle? Tell us in the comments how you currently recognize AI-generated images.
