Image-creating AIs are on the rise. In addition to the well-known Dalle-E 2 and Midjourney, many new services offer in one form or another to automatically generate graphic creations with a few words blown by Internet users. Stable Diffusion, for example, relies on its open source side, while TikTok embeds a similar tool to improve everyone’s creations.
Getty no longer wants AI
Despite their undeniable success, these heaps of algorithms trained to ingest millions of images are not to everyone’s taste. The famous site Getty Images, which hosts millions of illustrations, has decided to ban graphic creations generated by AIs because of the creative and legislative vagueness that surrounds them. In a statement to the site The VergeCraig Peters, CEO of Getty, says there is “genuine copyright concerns resulting from running these programs”. To avoid being at odds with the very real creators, Getty therefore preferred to take the lead in simply prohibiting the publication of such images on its platform.
Getty is arguably the biggest photo-sharing site to announce this policy change, but it’s not the first. Furaffinity.net, a site dedicated to producing anthropomorphic animal images (furry art), also banned the posting of AI-generated images a few weeks ago. The same goes for Inkblot and Newgrounds sites that host their members’ creations. The essential site DeviantArt has not yet announced anything on the subject, but the discontent of its members begins to rise.
A matter of plagiarism
According to the North American authorities, AI images are not protected by copyright, and it is therefore difficult to discriminate according to this criterion. But the problem could lie further upstream. All these AIs have been trained with millions of images available on the net, drawing in the process the creations of certain artists. Result, many images created by AI imitate the style of these creators without asking their opinion. It is therefore not really a problem of copyright, but of plagiarism on a large scale.
It is no coincidence that the databases used by these AIs are fond of images posted by digital artists. Indeed, these creations are very often accompanied by alternative text or legend describing what is represented. A definite advantage that allows the algorithms to understand each of the elements of the image without necessarily having to call on humans. The race for AI-generated images is obviously likely to be complicated as Midjourney, Dall-E and the others are good at creating photos similar to what any artist would do. But symbolically, the border has been drawn.