OpenAI has recently announced that they have embedded metadata into images generated by their Dall-E 3 system in order to indicate that they were created using AI technology. This metadata adheres to the C2PA industry standard, which is widely used in the field of photography. Users can now verify whether an image was generated by Dall-E or not by cross-referencing it with the Content Credentials website. Furthermore, Meta, the parent company of Facebook, has also announced that they will read and display this metadata to indicate that an image was created by AI.
OpenAI has emphasized that this type of metadata can be easily removed, unlike the watermarking system that Google has implemented, which is embedded within the image itself and cannot be separated by humans. Uploading images to certain social media platforms can potentially result in the extraction of this metadata.
According to OpenAI, the inclusion of metadata slightly increases the file size of the image, by approximately 3%, when generated via the API. However, if one were to order image creation through ChatGPT, the file size could increase up to 32%.
In summary, OpenAI has integrated metadata into images produced by Dall-E 3 to indicate their AI origin. This metadata follows the C2PA standard and can be cross-referenced on the Content Credentials website. However, it should be noted that this metadata can be easily removed and uploading images to specific social media platforms could result in its extraction. The inclusion of metadata may slightly increase the file size of API-generated images by about 3%, while images created through ChatGPT could experience a significantly larger file size increase of up to 32%.
Leave a Comment