AI-generated stickers in WhatsApp are already out of control

Meta will soon allow users to generate their own stickers using AI on WhatsApp, Instagram and Messenger. But it didn’t take long for them to already be hijacked by users.

The Meta Connect 2023 conference which took place on September 27 was an opportunity for Meta to unveil its own generative AI, Emu (Expressive Media Universe), which allows you to generate images using text prompts. The template can also be used to create completely personalized stickers to integrate into your WhatsApp, Instagram and Messenger conversations. And it didn’t take long for their use to be diverted.

When Waluigi and Winnie the Pooh carry guns

Some users have already been able to experiment with Emu’s engine for generating stickers in Meta applications. And given the many questionable creations that we can already see flourishing on X (Twitter), it would seem that the moderation and filtering tools are not the most effective.

Knowing that you can generate stickers from any assembly of words, some users have arrived at results that are questionable to say the least. Relayed by the user Pioldes on X (Twitter), some of these stickers represent well-known characters in a roundabout way: Waluigi carrying a rifle, Mickey Mouse smoking a joint and even Alex Jones kissing a dog.

Asked by the VentureBeat site about these first questionable creations, Andy Stone, communications director at Meta, assured that such moderation work was obviously an integral part of this type of tool: “ As with all generative AI systems, models may render inaccurate or inappropriate results. We will continue to improve these features as they evolve and users share feedback.”

On the one hand, these stickers use copyrighted characters like Mickey, Shrek or even Winnie the Pooh. On the other hand, these characters are diverted in a way that would absolutely not be to the taste of the rights holders.

And if we are to believe Meta’s official announcement on September 27, the firm wants to ensure that its ” templates are safer and more inclusive, provide safeguards for inappropriate conversations, and share templates publicly so people better understand how they work “. We are far from the mark at the moment.