Despite claims of ethical artificial intelligence (AI), the watermark Samsung adds to AI-edited photos can be effortlessly removed using another AI tool.
Samsung assures it will "apply the principles of social and ethical responsibility to AI". However, easily removable watermarks have rendered the company's ethical AI claims pointless.
When a Galaxy S24 user edits an image using the Generative edit feature, it automatically gets a watermark in the bottom left corner. This was part of Samsung's attempt to restrict users from spreading AI-altered or completely fake images online.
Much to Samsung's chagrin, X user Razar_the_Raven and the folks at Gizmodo have shown how easily you can edit an altered image to get rid of the double-star watermark. Once removed, the watermark doesn't reappear.
You can easily remove Samsung's AI watermark by simply object-erasing it 🤣 pic.twitter.com/YSVTI3avKt
— Raven (@Razar_the_Raven) January 18, 2024
The watermark, which is supposed to signify an image was generated using AI, can be removed using Samsung's own AI-based object erase tool, which the company says can eliminate and replace unwanted parts of a photo.
In Gizmodo's tests, some images where they used the object erase tool did not have a watermark, while others, including an AI-edited wallpaper, did. In the recordings of Samsung's Unpacked conference, the watermark appears when YouTuber MrBeast uses the object erase tool to remove the crane from his chocolate bar.
Will Samsung fix this?
It is still unclear whether Samsung is aware of these issues and if they are working on a fix. However, in a text that appears at the bottom of the abovementioned clip, Samsung noted that it is not taking the blame for any inaccurate deepfake.
"Generative Edit requires a network connection and Samsung Account login. Editing with Generative Edit results in a resized photo up to 12MP. A visible watermark is overlaid on the image output upon saving in order to indicate that the image is generated by AI. The accuracy and reliability of the generated output is not guaranteed," the Korean brand wrote.
Just like Google's Magic Eraser, the object erase tool replaces the pixels of unwanted objects in an image and blends them with the background. However, the system does not replace the same watermark in the new image for unknown reasons.
Major players in the AI space like Adobe and even the White House suggest watermarking is an effective way when it comes to restricting the spread of AI deepfakes online. These watermarks add metadata to each image to declare the image was modified.
Regretablly, a person with a little know-how can edit an image's metadata without breaking a sweat. As if that weren't enough, the Gizmodo report notes that users flipping mindlessly through Instagram aren't even likely to notice the metadata. Nevertheless, even if Samsung makes watermarks more impervious, most users will use some other apps to remove them.
Someone on reddit has posted an apparent Samsung Croatia advert
— Holly - I like tech (@AnxiousHolly) January 13, 2024
it says the new Galaxy AI features on the Galaxy S24 series will be free of charge until 2025…
so does this mean Samsung will charge its users to use the new Galaxy AI features? 😬💀
Yikes! 😟 pic.twitter.com/q22flLHSrd
Meanwhile, hawk-eyed Samsung fans have found evidence that suggests most of these Galaxy S24 AI features will be free only until 2025.
While it will be interesting to see whether Galaxy S24 users will shell out extra to use these AI-powered tools, a separate report claims Galaxy S24 AI will initially not be available in all languages, which is another bummer.
