By February 12th, OpenAI will introduce new watermarks in the images generated through FROM-E 3.
Following the directives of Coalition for Content Provenance and Authenticity (C2PA), any type of content generated by the website and fromAPI of DALL-E 3, will be equipped with invisible metadata and a symbol, with the latter clearly visible at the top left of each image created.
According to what OpenAI stated, this introduction will not influence the times and quality of image generation even if, for obvious reasons, their dimensions may be slightly larger than in the past. This introduction certainly does not represent anything new for the OpenAI context given that the same company has already tried to adopt similar solutions also with regard to the text generated with ChatGPT.
The aforementioned C2PA is a group of tech giants that promotes the use of watermarking and similar tools to verify the provenance of AI-generated content. In this sense, Adobe provided a logo that serves as a standard for quickly identifying what is artificially created from what is human-made.
OpenAI and beyond: but can watermarks really avoid abuse?
A very similar move was also made by Metawhich recently announced the addition of tags to AI-generated content across its platforms.
This move is very useful to avoid abuse of such tools, even if watermarking is not a permanent solution to the problem. According to OpenAI, in fact, C2PA metadata can be easily manipulated or removed, even unintentionally. A simple passage of the offending image, through a screenshot and a crop of it, is enough to eliminate any sign of labeling.
Despite this, the leading AI company seems optimistic”We believe that adopting these methods to establish provenance and encourage users to recognize these signals is critical to increasing the trustworthiness of digital information“.