Barely available, Midjourney is already wreaking havoc, with many Internet users having fun posting more realistic fake photos on social networks. Beware of misinformation!

Barely available, Midjourney is already wreaking havoc, with many Internet users having fun posting more realistic fake photos on social networks. Beware of misinformation!

Image-generating AIs fascinate Internet users as much as they worry. Indeed, they are able, in just a few seconds, to create incredible images with a wide variety of artistic styles from a simple text command. Surrealist painting, pencil portrait and, of course, realistic photo… Nothing is impossible! DALL-E, created by the developers of ChatGPT, and its rival Midjourney are among the artificial intelligences that offer the most beautiful renderings. The latter has also recently experienced a serious improvement with the release of its fifth version (v5), allowing it to generate much more realistic and detailed images (see our article). Also, it did not take long before Internet users shared their creations on social networks. And political figures are obviously an incredible source of inspiration.

Emmanuel Macron proudly enthroned on a pile of trash cans, in the middle of a horde of unleashed zombies, Donald Trump in an orange jumpsuit, Barack Obama and Angela Merkel sharing an accomplice moment, or even Vladimir Poutine and Joe Biden laughing… Internet users are not lacking in imagination! Anyone can create images that are sometimes impressively realistic, bordering on deepfakes – images that use artificial intelligence to superimpose a face on another face, and therefore reproduce “fake” people -, which are feared risks of manipulation of information…

Midjourney: strikingly realistic caricatures

In response to the 49.3 used to force through the much-maligned pension reform in the National Assembly, as well as the resulting strikes and demonstrations, the creations taking Emmanuel Macron as their target – or source of inspiration, it all depends on the point of view – are invading Twitter. The one we nicknamed Jupiter is staged in various situations: as Napoleon, sipping a drink at a bar, as a zombie, in full demonstration or even pensive, wandering in the streets in the middle of garbage cans and burnt barricades. .. And, of course, proudly enthroned on a pile of garbage cans, an explicit reference to the garbage collectors’ strike and the streets strewn with waste in several districts of Paris. On the other hand, the AI ​​seems to have more difficulty imitating Prime Minister Élisabeth Borne – it must certainly be missing images in its database used for its training.

Politicians from other countries are no exception either. In the United States, these web cartoons revolve around the hypothetical arrest of Donald Trump, who called on his supporters to demonstrate. We therefore see him chased by the police in the street, surrounded by police officers who arrest him, then imprisoned with the famous orange jumpsuit. The rivalry between Joe Biden and Vladimir Putin is also regularly staged.

Midjourney: from laughter to disinformation, there is only one step

Even if the rendering is not yet perfect – we can note some texture problems, faces that don’t fit well with the rest of the body, silhouettes that stand out too clearly from the background or even faces that are a little frozen –, the progress shown in the space of a few months is simply impressive, especially in terms of light management with version 5 of Midjourney. And that is more than enough to fool inattentive Internet users, who would see the image pass on their news feed.

If Internet users specify in most cases that their images were produced by Midjourney and that the cartoonish style of the latter leaves no doubt about their falsity, it is easy to imagine that the technology could be used for the purposes of disinformation in a near future. On DALL-E, the question does not arise for the moment since the filters applied by OpenAI prevent the AI ​​from generating representations of public figures and sensitive subjects – in particular related to current events. But for the rest…

Midjourney: adapting to the proliferation of deepfakes

Twitter is undoubtedly one of the most popular sharing platforms for these images. In its rules, the blue bird declares to want to fight against “manipulated media” and reserves the right to place a warning in the event that they are intended to mislead. It remains to be seen what this gives in practice, especially with the famous “freedom of expression” so dear to Elon Musk. For its part, TikTok has just updated its rules to regulate deepfakes and other false images, adding a new branch dedicated to “synthetic and manipulated media”. The social network recommends using a “synthetic”, “fake”, “unreal” or “modified” sticker or caption for better transparency with users. In addition, it completely prohibits deepfakes on private persons, but makes an exception for “public figures with an important public role”as “a politician, a business leader [ou] a celebrity”. Videos using AI to imitate them will still be able to be posted, unless “the content is used for promotional purposes or violates any other policy”.

Image-generating AIs represent a real challenge for the media, who will have to be extra vigilant to avoid being misled by images created from scratch. A real nightmare for fact-checking! It will be necessary more than ever to doubt what we see on the Internet!