The Chinese social network has implemented a new effect called “AI greenscreen” allowing everyone to create their own little work of art. Nothing could be simpler: just have an idea and type it in the text box, the artificial intelligence will then generate an image that can be used in the background of a video. While this new feature is strongly reminiscent of DALL-E 2, it is nevertheless much more basic than other “text-to-image” software.
Although the new TikTok filter offers some creative freedom, it remains limited compared to other artificial intelligences. It should first be noted that all the images generated are actually sort of abstract drawings and that a realistic photo image will never be generated. Producing precise and complex drawings requires a lot of computing power and a fairly substantial technological need. Given that it is only one filter among hundreds of others, the investment would undoubtedly be disproportionate for TikTok.
Added to this is the disproportionate number of users of the application. Allowing more than a billion individuals to easily generate realistic photo images would present a certain danger when we know the excesses that the Internet can have. In their article on the subject, The Verge attempted to have artificial intelligence generate obscene or violent images, to no avail. The abstract aspect of each image allows moderation in the very creation of the images.
For example, in depictions of the assassination of political figures, no face is recognizable. The same applies to images of nudity where no bodies or obscene parts are distinguishable, even if the color palette matches that of the skin. When we see the proliferation of these image-generating artificial intelligences, we understand that the social network, often at the heart of viral phenomena on the Internet, also wants to take advantage of the growing popularity of these AIs.