Microsoft announces attitude to watermark all AI art by its own products

At the developer conference 'Microsoft Build', Microsoft will put a 'digital watermark' that indicates that it was created with AI in all AI art by its own products that use image generation AI, such as Bing Image Creator and Microsoft Designer . clarifying.

Microsoft Build 2023 Book of News
https://news.microsoft.com/build-2023-book-of-news/

Microsoft will ID its AI art with a hidden watermark | PCWorld
https://www.pcworld.com/article/1923811

Adobe, Arm, Intel, Microsoft, and Truepic have established the `` Coalition for Content Provenance and Authenticity (C2PA) '', an organization that will develop open standards to indicate the origin of digital images and whether images are made by AI in 2021. .

For example, the image generation AI ' Firefly ' announced by Adobe has a function called 'Content Credentials' that embeds digital watermarks in metadata, but this is only supported within the Adobe app. , the metadata visible in other apps, such as the Windows 11 standard viewer, is not reflected.

However, by adopting digital watermarking in the C2PA standard, consumers will see a small mark indicating that it is an AI work on the top and side of the generated image in addition to the metadata, so it will not be affected by the application or environment. Electronic watermarking is realized.

At the time of writing the article, the Bing logo is drawn small in the lower left as a sign that it is AI generated in the image created with Bing Image Creator. However, the images generated by Microsoft Designer did not have such markings.

This time, Microsoft says that the images generated by both tools will be signed according to the C2PA standard, and the image metadata will clearly disclose the provenance. Microsoft said, ``We will sign and verify AI-generated content based on the C2PA standard. Within 2023, Microsoft will comply with the C2PA standard for key images and videos in Microsoft Designer and Bing Image Creator.'' I'm here.

In addition, Microsoft has announced ' Azure AI Content Safety ', a service that AI supports content moderators. Originally, it is the role of humans to check content for offensive or inappropriate elements, but Azure AI Content Safety is a service in which AI takes over this work, and is scheduled to start in June 2023. I'm here. The service price is $ 1.50 (about 200 yen) per 1000 image scans.

Microsoft says it can simplify the content checking process by taking the lead in adopting the C2PA standard and encouraging others to adopt it. For example, for camera manufacturers such as Leica and Nikon, by incorporating the C2PA standard into the camera, there is also a mechanism to prove that the image is not an image generated by AI, but a real photo actually taken with the camera. It seems that there is