Google DeepMind has unveiled Synth ID, a system that embeds imperceptible watermarks into AI-generated images to verify that they have not been created by humans, preventing the problem of fake and manipulated images in real events. This feature was announced at Google I/O 2023 and has now been developed into Synth ID by the DeepMind team in collaboration with Google Research.
Synth ID is a method of embedding watermarks that are imperceptible to the human eye and difficult to remove. Simple image modifications such as applying filters, changing colors, compressing image files, or cropping images to make them smaller cannot eliminate the watermark of Synth ID (although extreme image manipulation can still destroy it).
The background of Synth ID is based on deep learning techniques for both watermarking and identifying watermarks to find the best way to make them invisible and undetectable by humans, while still being checkable by Google’s tools.
Initially, Synth ID will be used with Imagen, Google Cloud’s AI-based image creation service, in a limited scope before gradually expanding to other services within the Google ecosystem.