Would you believe this was probably the safest image we could have used here? | Image: Meta / @Pioldes
Some early user tests for Meta’s new AI-generated sticker tool have resulted in some dubious (and rather hilarious) creations. After gaining access to the new AI-generated sticker tool on Facebook Messenger, X user @Pioldes was able to create a host of inappropriate sticker images — including child soldiers, gun-wielding Nintendo characters, Mickey Mouse taking a crap, and nude illustrations of Canadian prime minister Justin Trudeau.
Other examples show that Meta’s AI-sticker tool will also happily slap a pair of breasts onto…well, just about anything, judging by the busty images of Sonic the Hedgehog and Karl Marx. There’s even a sticker that depicts a woman breastfeeding Pikachu.
found out that facebook messenger has ai generated stickers now and I don’t think anyone involved has thought anything through pic.twitter.com/co987cRhyu
— podesbiens.bsky.social (@Pioldes) October 3, 2023
The AI-generated chat stickers were announced last week at Meta’s Connect event, alongside a new AI image editor for Instagram. Powered by Meta’s Llama 2 large language model — the company’s ChatGPT rival — AI-generated stickers allow users to create “multiple unique, high-quality stickers in seconds” using text-based prompts. AI-generated stickers are currently rolling out to “select English language users” for Facebook Stories, Instagram Stories and DMs, Messenger, and WhatsApp over this month, so it’s unclear how many users currently have access to the feature.
I’m still trying to figure out if I love this or hate this pic.twitter.com/rrm6oKyVQb
— Kazzer✘ (@kazzerx) October 4, 2023
Certain words do seem to be blocked and using them in prompts will warn the user that their description might violate Meta’s community guidelines. However, other X users have reported that you can still generate inappropriate content using typos or descriptions of restricted words. In some cases, prompts like “World Trade Center” generated problematic images without any additional descriptors at all.
LPT: typos of restricted words will eventually generate the prohibited content! pic.twitter.com/6PnrTJ3U7x
— Buttered Jorts (@ajn142) October 4, 2023
This sort of tomfoolery is common with the launch of AI tools and likely why Meta is pursuing a limited rollout of the AI-generated sticker feature. That way it can address and correct abuse before it spreads to the masses. We have reached out to Meta for comment and will update this story if we hear back.
