Meta launches Muse Image, an image model

Meta Superintelligence Labs, the unit led by Alexandr Wang, released its first image-generation model on July 7. Muse Image is an agentic system, it can search the web, run code, and self-refine its own output, and Meta says it currently ranks second on Arena’s text-to-image and editing leaderboards, behind GPT Image 2. A companion video model, Muse Video, is coming soon and already ranks third on Arena’s video leaderboard.

The feature drawing the most attention isn’t image quality: users can @-mention any public Instagram account inside a Muse Image prompt, and the model will pull that person’s real likeness from their public photos into a generated scene. It’s live now in the Meta AI app, in Instagram Stories in the US, and on WhatsApp in a limited set of countries, free to use. Meta says generated images carry an invisible “Content Seal” watermark and that the photo-pulling feature is opt-out, a public account is included by default unless its owner turns it off in settings.

That opt-out default is the part drawing criticism. Meta’s approach treats existing public photos as a standing input to a generative system rather than something each person actively agrees to share, and outlets including Wired flagged that most users won’t know the setting exists.

If you build anything that touches Instagram’s platform or public profile data, expect this policy, not the model’s capabilities, to be the thing regulators and users push back on first.