OpenAI recently announced its decision to enhance transparency around its AI-generated images produced by DALL-E 3 by adding watermarks to their metadata. The company stated that it will utilize the open technical standard endorsed by the Coalition for Content Provenance and Authenticity (C2PA), incorporating details such as the image’s AI-generated nature, the specific AI tool used (in this case, DALL-E 3), and the application utilized to create the image.
This initiative comes in response to Meta’s call for AI firms to adopt a standardized approach to aid detection tools in identifying and labeling AI-generated content on social media platforms.
In a detailed post, OpenAI outlined the implementation process and technical specifics of the watermarking procedure. Images generated using ChatGPT on both the Web client and API, leveraging the DALL-E 3 model, will now feature a new metadata format compliant with the C2PA standard. This watermarking process will also extend to the ChatGPT app by February 12. The C2PA standard employs watermarking technology to embed information within the image itself, visible as a CR symbol in the top left corner of the image, with more detailed information accessible via the metadata.
Through this metadata, users can ascertain the origins of the image, including details about the AI model and the specific application utilized. OpenAI provided examples demonstrating that the metadata includes a content summary indicating that the image was created using an AI tool, along with information on the tool and the underlying AI model used. The company assured that adding this metadata may marginally increase the image’s size without compromising quality.
However, while this approach enhances image security compared to a mere visual marker, it is not foolproof. OpenAI acknowledged that many social media platforms strip metadata from uploaded images, and capturing a screenshot of the image would also remove it. Hence, this method may not always suffice to confirm whether an image was generated by DALL-E 3 or other AI models.
The C2PA initiative boasts participation from prominent companies such as Adobe, Microsoft, BBC, Sony, Leica, Nikon, and others, advocating for the widespread adoption of this technology to accurately detect and label AI-generated content. The CR symbol, developed by Adobe, is part of this collective effort.
Leave a Reply