OpenAI's image generator DALL-E3 will add watermarks to image metadata as more companies roll out support for the Content Authenticity Alliance (C2PA) standard. The company said C2PA’s watermark will appear in images generated by the ChatGPT website and in the DALL-E3 model’s API. Mobile users will get the watermark until February 12th. The watermark will include an invisible metadata component and a visible CR symbol, which will appear in the upper left corner of each image.
People can verify the provenance (i.e. which AI tool was used to create the content) of any image generated by the OpenAI platform through sites such as Content Credentials Verify. So far, only static images can be watermarked, not videos or text.
OpenAI says adding watermark metadata to images "has a negligible impact on latency and does not affect the quality of the resulting image." It also slightly increases image size for some tasks.
The C2PA group, made up of companies including Adobe and Microsoft, has been pushing for the use of a "content certificate" watermark to identify the provenance of content and show whether it was produced by humans or artificial intelligence. Adobe has created a content voucher symbol that OpenAI is adding to the creation of DALL-E3. Meta recently announced that it will add tags to AI-generated content on its social media platform.
Identifying AI-generated content is one of the key directives of the Biden administration’s executive order on artificial intelligence. But watermarks are not a surefire way to stop misinformation. OpenAI pointed out that C2PA's metadata "can easily be deleted accidentally or intentionally," especially since most social media platforms usually delete metadata from uploaded content, and screenshots will also omit metadata from images.
"We believe that employing these methods to determine provenance and encourage users to recognize these signals is key to increasing the trustworthiness of digital information," OpenAI says on its website.