Le générateur d’images par IA de Meta s’émancipe vers une filigrane invisible de traçabilité

Mark Zuckerberg’s group has launched an autonomous website for its AI Imagine. This AI image generator, a competitor to DALL-E, Midjourney, or Stable Diffusion, was previously only accessible in Facebook, Instagram, and WhatsApp messenger. Based on Meta’s Emu language model, trained on over a billion publicly shared Facebook and Instagram contents, this free tool creates four different results from each textual description (quickly). They are generated in JPEG format. Available at imagine.meta.comit is currently only accessible to users based in the United States (or via a VPN tool) and requires a Meta account.

“We’ve enjoyed hearing people’s testimonials about how they use Imagine, the Meta AI image generation feature, to create fun and creative content in discussions,” said the Meta group about the beta version of this feature.

It should be noted that each image integrates a watermark in the lower left corner indicating that it is the result of Meta’s AI. The group says it intends to soon test an “invisible” watermark “resistant to common image manipulations such as cropping, color adjustments (brightness, contrast, etc.), screenshots, etc.”