The DALL-E 3 image generator from OpenAI will add new watermarks to image metadata using the C2PA specifications.
OpenAI shared this information through X, here is the post.
The company says that images generated on the ChatGPT website and the API for the DALL-E 3 model will have C2PA watermarks. Mobile users will also get the update by Feb 12th.
The company is saying that adding watermark metadata to images will have a negligible effect on latency and will not affect the quality of the image generation. Also, the image size will be increased slightly for some tasks.
The C2PA is a technical standard used by companies like Microsoft, Adobe, BBC, and others, which stands for Coalition for Content Provenance and Authenticity. This standard was created to address the prevalence of deepfakes and misinformation “by certifying the source and history (or provenance) of media content.”
Meta recently announced that they will start labeling AI-generated images from any tool on Facebook, Instagram, and Threads. Identifying AI-generated content is the top priority for US government and tech companies. But only watermarking may not have an impact we are hoping it would.
The C2PA specifications metadata can also be removed through software. And most of the time, social media platforms remove the metadata when users upload an image.
OpenAI says “We believe that adopting these methods for establishing provenance and encouraging users to recognize these signals are key to increasing the trustworthiness of digital information,”
Only time will tell whether this will be enough to identify AI-generated images on social media or not.
