WhatsApp starts testing AI-generated stickers

Meta-owned WhatsApp has started testing a new AI-generated stickers feature. WABetaInfo reports that some testers in the Android WhatsApp beta program have spotted the new AI-powered stickers, allowing WhatsApp users to generate a sticker based on a text description.

Android Police points out that it’s not clear which generative AI model WhatsApp has picked for this feature, as WABetaInfo’s report only notes that the stickers are generated using “a secure technology offered by Meta.”

The feature looks like it will work similarly to how Midjoureny or OpenAI’s DALL-E models generate images based on text alone. It seems you’ll be able to use this WhatsApp feature to create a basic and personalized image that can be shared as a sticker with WhatsApp friends or groups.

With any generative AI model there will be concerns over what images and stickers could be generated and shared easily on WhatsApp, particularly as WhatsApp has previously had to place limits on the forwarding of viral messages in the past after the spread of misleading images during elections, mob attacks and killings in India, and misinformation related to covid-19. WhatsApp has the ability to report any inappropriate or harmful stickers, but it’s not clear what additional protections are in place on the AI model side here.

WABetaInfo says stickers that are generated using this new AI-powered feature are “easily recognizable,” which suggests there could also be some form of watermark or indication that these stickers were created using generative AI models. Microsoft uses OpenAI’s DALL-E model to allow Bing Chat users to create images, but the company places a clear Bing label on images. Instagram is also reportedly working on labels that would identify when AI models have been used to create content.