Google Adopts Metadata Standards for AI-Manipulated Images

Google Search Central, in its updated documentation, reveals that Google can now label images that are developed or manipulated by artificial intelligence (AI). 

Furthermore, Google has also removed the “AI-generated” metadata from beta status because that label is now fully supported by Search.

 

Image Metadata on Google Images

Google's documentation reveals that when you specify image metadata, Google Images will show more details about who created the image, how users can use it, and related credit information.

For example, when an image has licensing information, it will receive a "Licensable" label, include a license link, and provide further details on how users can use the image according to its license.

Regarding image metadata standards, the International Press Telecommunications Council (IPTC) is responsible for setting them.

Image metadata is important for providing image descriptions and copyright information. Although IPTC is an international press standards organization, the metadata standards they develop are now used by Google Images in addition to Google News.

 

Image Metadata and AI Manipulation

Google has adopted metadata standards for images manipulated by artificial intelligence, including Convolutional Neural Networks (CNN) and Generative Adversarial Networks (GAN).

In this latest metadata update, there are two types of AI image manipulations supported by Google: inpainting and outpainting.

Inpainting is the process of reconstructing or restoring an image by filling in missing parts. This process also includes algorithmic manipulations that add elements to the image.

Outpainting is an algorithmic process that adds assets to an image beyond what is in the original and extends it beyond the original boundaries.

For both types of manipulations, Google can support their metadata with compositeWithTrainedAlgorithmicMedia.

However, this property is only available in beta versions and IPTC image metadata. Adding digital source properties such as compositeWithTrainedAlgorithmicMedia, compositeSynthetic, and trainedAlgorithmicMedia makes your images eligible for the AI label, although it may not be immediately visible to users as it is still under development.