The latest Galaxy S24 series introduces widespread AI photo editing, following the Pixel 8 and Android 14. Despite my disapproval of artificial enhancements, Samsung addresses potential criticism by ensuring all AI-edited Galaxy photos display watermarks and metadata, confirming manipulation.
The concept is commendable, but it falters when you discover the watermark can be entirely eliminated, as demonstrated in tests with the device. Even if Samsung addresses the problem through a software update, this visual marker can still be circumvented.
I’ve been critical of Google’s Magic Editor since its introduction last spring. While it’s impressive that generative AI allows photo manipulation without Photoshop expertise, the act of faking photos has been present since the inception of photography.
However, the instant manipulation by AI democratizes the creation of fakes, presenting a significant potential for abuse in our current world.
Some individuals may extend beyond altering weather or rearranging elements to present a certain event more favorably in photos. They manipulate images with the goal of influencing the viewer. While this was feasible before generative AI on phones, it has now become remarkably simple.
The Galaxy AI incorporates a feature called Generative Edit, allowing users to adjust picture edges or relocate subjects, akin to Google’s Magic Editor. Following such alterations, Samsung appends a watermark in the lower left corner, along with metadata, signaling that the image underwent AI-based editing.
Samsung employs a three-star watermark across the Galaxy S24 as the universal symbol for AI features. To activate AI, simply tap the icon. If the watermark appears in photos, it signifies artificial manipulation. However, a significant drawback is that the watermark can be effortlessly eliminated using the same phone.
As demonstrated in the video above, the watermark is removable—unless, of course, the video itself is fabricated. This highlights the challenge posed by generative AI in media: everything seen cannot be blindly trusted.
Gizmodo successfully duplicated the experiment and discovered that not all images created using the object erase tool displayed the watermark initially.
There’s a possibility that Samsung is cognizant of this issue. As highlighted by Gizmodo, during the Mr. Beast segment of the Unpacked event, Samsung included a disclaimer cautioning about potential AI-related issues. Mr. Beast utilized Galaxy AI capabilities to eliminate a crane from a photo (refer to the video below), prompting Samsung to issue the disclaimer
Samsung has the option to address this software issue and potentially make the watermark permanent. It’s important to note that removing the watermark doesn’t eliminate the metadata, although most users tend to overlook this information. Additionally, changing metadata is a relatively simple task.
It’s worth mentioning the obvious – even if Samsung successfully resolves the bug, one can simply capture a screenshot of the manipulated AI image and crop it to remove the watermark. In essence, relying solely on watermarks doesn’t effectively address the issue of fake AI photos.
Samsung, along with other providers of advanced AI-powered photo-editing tools, should explore more effective methods to alert users to the artificial manipulation of images or clips through AI, especially for those who may be unsuspecting.
