Google’s AI arm, DeepMind, has developed SynthID to identify machine-generated images.

By embedding changes to individual pixels in images, watermarks are invisible to the human eye, but detectable by computers.

However, DeepMind said it is not foolproof against extreme manipulation.

It is becoming increasingly difficult to distinguish between real and artificially generated images as technology advances.

With more than 14.5 million users, Midjourney is one of the most popular AI image generators.

With simple text instructions, people can create images in seconds, raising questions about ownership and copyright.

Google has its own image generator called Imagen, and its watermarking system only applies to images created with this tool. A watermark is typically a logo or text that is added to an image to show ownership as well as to prevent the image from being copied or used without permission.

In the bottom-left corner of images used on the News website, there is usually a copyright watermark.

Since these kinds of watermarks can easily be edited or cropped out, they are not suitable for identifying Al-generated images.

Tech companies use hashing to create digital fingerprints of known videos of abuse, so they can spot them and remove them quickly if they spread online. If the video is cropped or edited, these can also become corrupted.

Google’s system creates a watermark that is effectively invisible, allowing users to use its software to instantly determine whether a picture is real or not. In an interview with the , DeepMind’s head of research, Pushmeet Kohli, said its system modifies images so subtly that “to you and me, it doesn’t look any different”.

Even after an image is cropped or edited, the firm’s software can still detect the watermark, unlike hashing.

You can change the colour, the contrast, even resize it… [and DeepMind will still see that it is AI-generated], he said.

The company needs people to use the system to learn more about how robust it is as this is an “experimental launch.”