Microsoft has chosen its own format for labeling images created with the help of artificial intelligence: “watermarks” are superimposed on images in the Paint and Photos apps and contain information that indicates the user who sent the request to the artificial intelligence.
Image source: BoliviaInteligente / unsplash.com
The label was introduced in compliance with the requirements of new EU transparency regulations. They recommend that AI-generated content should be tagged in a machine-readable format and that associated metadata should at least indicate whether the content was processed by AI or generated entirely by AI. The exact method has not yet been defined, and Microsoft decided to expand this information – as a founding member of the Content Provenance and Authenticity Alliance (C2PA), the company decided to implement tagging in the Paint and Photos apps and use a globally unique identifier (GUID) associated with the query that generated the image.
The invisible GUID watermark is a 16-byte integer and is different from the visible watermark provided by Microsoft to users of Paint and Microsoft 365 AI features. By understanding the GUID query associated with the image, the company has the technical ability to identify the user who issued the query. It is worth noting that when the request is generated, Paint will also transmit the value of the previous PromptGenerationId, which means that Microsoft can also connect the request sequence.
Anthropic, OpenAI, and Google have launched AI content tags to meet AI needs. Only models with open scale (including stable diffusion) can work without markers; the most reliable way to maintain confidentiality is to run the model locally on the user’s device rather than in the cloud.
If you find an error, select it with your mouse and press CTRL+ENTER.
