OpenAI has announced that they will embed metadata in images generated by Dall-E 3 to indicate that they were created using AI. The metadata embedded in the images adheres to the C2PA standard, which is commonly used in the photography industry. Users can cross-check the images with the Content Credentials website to verify if they were created using Dall-E. In a similar vein, Meta has also announced that they will read this set of metadata and label the images as being AI-generated.
However, OpenAI warns that this type of metadata embedding can be easily removed, unlike the watermarking system that Google has implemented, which is embedded within the image and cannot be deciphered by humans. Uploading images to certain social media platforms already allows for the extraction of metadata.
OpenAI further explains that the inclusion of metadata will slightly increase the file size of the image (approximately 3%), if the image was generated via the API. However, if the image was generated through ChatGPT, the file size could increase by up to 32%.
TLDR: OpenAI will embed metadata in Dall-E 3 generated images to indicate their AI origin. Metadata adheres to C2PA standards and can be verified using the Content Credentials website. Meta will also label images as AI-generated. However, this metadata can be easily removed, unlike Google’s watermarking system. File sizes may increase slightly when metadata is included.
Leave a Comment