SynthID - Google’s new invisible watermark technology for AI art

Google has taken a step towards transparency in AI-generated images. Google’s DeepMind announced a watermarking (watermark) or identification tool for AI generated art, called SynthID. Unlike regular visible watermarks, which can be easily removed or altered, SynthID leaves no trace on the surface of the image itself. Instead, it embeds a hidden code within the pixels of the artwork, acting as a secret identifier that remains undetectable to the human eye.

In partnership with Google Cloud, SynthID will first be available to “a limited number” of customers. For now, It is exclusively available for Vertex AI customers using Imagen.

To clear the fog, With Vertex AI, your teams can collaborate using a common toolset and scale your applications. It’s done using Google Cloud’s benefits. And Imagen is Google’s AI art generator.

With the rapid evolution of generative AI technologies, computer generated imagery, also known as ‘synthetic imagery’, is becoming more difficult to distinguish from non-AI imagery.

Despite its huge potential, generative AI also poses new risks, such as enabling creators to spread false information, whether intentionally or unintentionally. To prevent the spread of misinformation, people need to be able to identify AI-generated content when they’re interacting with it.

We have all seen the AI – generated posts of the Pope’s trendy outfit. One more such as the arrest of former U.S president Donald Trump. It became quite viral. It doesn’t take much long for a funny representation to turn into a political war which will do more damage to the society.

What measures will be taken?

Following a July White House meeting, seven AI companies agreed to develop a voluntary commitment to watermark audio and visual content to identify AI-generated content. The first company to launch such a system is Google.

With SynthID, Vertex AI customers can generate AI-generated images responsibly and identify them with confidence. Google says, Although this isn’t a foolproof system, their internal testing shows that this technology is accurate against many common manipulations of images.

SynthID combines watermarking and identification to get the job done.

Imagen is likely responsible for part of the image if a digital watermark is detected. This tool provides three confidence levels for interpreting the results of watermark identification.

Metadata is one of the most widely used methods for identifying digital content. It provides information such as Who created it and when was it created and such.

Metadata is stored with the image file. Digital signatures added to metadata can indicate whether or not an image has been modified. A file can be easily identified when the metadata information is intact. However, metadata can be manually removed or even lost when files are edited.

SynthID’s watermark is embedded in the pixels of an image, so it is compatible with other image identification approaches that rely on metadata. It remains detectable even when metadata has been lost.