After announcing during Google I/O 2023 that it would add recognition tags to images generated by automatically generated artificial intelligence, Google announced at the Google Cloud Next '23 event that Google Cloud and Google DeepMind will collaborate toA mechanism called "SynthID", adding identification information that cannot be distinguished by the human eye to the automatically generated image content.
This mechanism is currently being tested in beta form, and Google has not yet confirmed when it will be actually implemented, but it is expected to serve as a tool to assist in determining whether image content is post-produced or has been manually digitized.
According to the description, "SynthID" adds a layer to the image that is invisible to the human eye but can be interpreted by the system. This layer does not affect the image quality, resolution, or compression ratio, and even allows the image to be opened and used normally. Furthermore, the added layer information does not write the image metadata, but is directly embedded in the original image pixels. Therefore, the system can determine whether the image content has been modified by checking whether the image metadata has been tampered with.
The judgment results are divided into three basic categories: images are automatically generated, not automatically generated, and possibly automatically generated. At the same time, the judgment is mainly based on analysis through Google's machine learning platform Vertex AI, combined with Google's text-to-image model Imagen.
However, judging from Google's announcement that it will connect more artificial intelligence model frameworks to Vertex AI, "SynthID" should be able to judge more image content generated by different artificial intelligence in the future.
Prior to this, Adobe had also proposed using artificial intelligence to determine whether image content was fake, and formed an alliance with Twitter and the New York Times to use a technology calledContent Authenticity Project (Content Authenticity Initiative) to combat digital fake news content.
