Google Search Central recently updated its documentation to include support for labeling images that have been altered or extended using AI. Additionally, Google has officially moved the “AI generated” metadata out of Beta status, signifying that the “AI Generated” label is now fully endorsed in search results.
IPTC Photo Metadata
The International Press Telecommunications Council (IPTC) is responsible for establishing standards, including those for photo metadata. Photo metadata provides essential information about a photograph, such as details regarding copyright, licensing, and image descriptions.
Although IPTC is primarily an international press standards organization, the metadata standards they develop are utilized by Google Images beyond the scope of Google News. This metadata enables Google Images to display additional information about each image.
According to Google’s documentation, implementing this metadata offers several advantages:
“By specifying image metadata, Google Images can present comprehensive details about the image, including attribution to the creator, usage rights, and licensing information. For instance, providing licensing details can qualify the image for a Licensable badge, which links to the license and provides further information on permissible uses.”
AI Image Manipulation Metadata
Google recently implemented metadata standards for images altered using AI algorithms commonly employed for image manipulation, such as convolutional neural networks (CNNs) and generative adversarial networks (GANs).
The updated metadata addresses two types of AI image manipulation:
- Inpainting
- Outpainting
Inpainting
Inpainting is typically understood as a technique to improve or reconstruct an image by filling in missing parts. However, inpainting can also refer to any algorithmic manipulation that adds new elements to an image.
Outpainting
Outpainting refers to the algorithmic process of expanding an image beyond its original boundaries, adding additional content that extends beyond the original photograph.
Google has introduced support for labeling images manipulated in both these ways using a new metadata attribute called “compositeWithTrainedAlgorithmicMedia” under Digital Source Type. This metadata is embedded directly into the digital image rather than being structured data according to Schema.
Here’s the excerpt from Google’s updated documentation:
“Digital Source Type
compositeWithTrainedAlgorithmicMedia: This indicates that the image is a composite created using trained algorithmic media, possibly through inpainting or outpainting operations.”
Label For “AI Generated” – algorithmicMedia Metadata
Google has also removed the Beta status from the algorithmicMedia metadata specifications, allowing images created with AI to be labeled as AI Generated when the algorithmicMedia metadata is embedded in the image.
Here is the original documentation prior to the update:
“algorithmicMedia: The image was created purely by an algorithm not based on any sampled training data (for example, an image created by software using a mathematical formula).
Beta: Currently, this property is in beta and only available for IPTC photo metadata. Adding this property makes your image eligible for display with an AI-generated label, but you may not see the label in Google Images right away, as we’re still actively developing it.”
The change made to the documentation involved removing the entire second paragraph, which mentioned the Beta status. Interestingly, this update was not reflected in Google’s changelog.
According to Google’s Search Central documentation changelog:
“Supporting a new IPTC digital source type
What: Added compositeWithTrainedAlgorithmicMedia to the IPTC photo metadata documentation.
Why: Google can now extract the compositeWithTrainedAlgorithmicMedia IPTC NewsCode.”
Original news from SearchEngineJournal