Upcoming Invisible Watermarks for DALL-E Generated Images on Azure OpenAI Service

OpenAI’s DALL-E models are capable of generating images from text prompts. The Azure OpenAI Service provides REST API access to OpenAI’s DALL-E models, enabling developers to programmatically create images within their applications. Today, Microsoft announced an innovative built-in feature within the Azure OpenAI Service known as ‘Watermarks.’ This new Watermarks feature will add invisible watermarks to all images produced using DALL·E, enhancing transparency and safeguarding AI-generated images.

The recent surge in image generation models has led to a notable increase in disinformation and the proliferation of AI-generated deepfakes. Thus, the ability to recognize AI-generated content has become crucial. The newly introduced invisible watermark, embedded in AI-generated images, can be detected using specialized tools, yet it remains undetectable to the naked eye, preserving the integrity of the image. Additionally, the watermark’s integrity holds even if the AI-generated image is resized or cropped.

The invisible watermarks will contain metadata about the image’s origin, represented through a manifest linked to the image. This manifest is cryptographically signed by a certificate that references Azure OpenAI Service and will include the following details:

“description” – This field indicates “AI Generated Image”for all images produced by DALL-E models, confirming the AI-generated nature of the image.
“softwareAgent” – This field denotes “Azure OpenAI DALL-E”for all images created by DALL-E series models integrated within the Azure OpenAI Service.
“when” – This timestamp marks when the Content Credentials were established, highlighting similarities with other Azure AI services.

This is not the first instance where Microsoft has implemented watermarks. Last year, Microsoft rolled out watermarks for voice outputs generated with Azure AI Speech’s personal voice feature, enabling users to discern whether the speech was synthesized through Azure AI Speech.

Furthermore, Microsoft is collaborating with other prominent AI industry players, including Adobe, Truepic, and the BBC, to ensure consistency in watermarking, cryptographic metadata, and detection techniques across various platforms.

This initiative by Microsoft represents a pivotal stride towards responsibly deploying AI technology and addressing the misuse of AI-generated content.

Source