Meta Shut Down an Instagram AI Tool That Could Generate Your Face Without Your Permission

Meta has disabled a feature in its new Muse Image AI model that let Instagram users create AI-generated pictures by tagging the accounts of real people — including strangers. The company removed the capability just days after introducing it on July 7, updating its original announcement to say the feature is "no longer available."

How the Feature Worked

Muse Image was Meta's new tool for generating images using AI. One of its features worked like this: if you typed a prompt and tagged an Instagram account by name, the AI would look at that person's publicly posted photos and use them as a reference to create new pictures. Those new images would carry the appearance, face, and style of the person you tagged.

The problem was simple: it worked on any public Instagram account, not just accounts belonging to the person creating the image. You could tag a stranger, a celebrity, even a public figure, and the AI would generate images of them. Account owners would have no way of knowing this was happening unless they dug into Instagram's settings themselves.

The Backlash

Within days, criticism emerged from actors' unions, privacy advocates, and users. The concern centered on a key issue: just because someone's photos are public does not mean those photos should be used to create AI-generated images of them without permission. The feature risked generating fake pictures of real people — essentially, AI-created deepfakes — and there was no consent involved.

Meta's initial response was to add an opt-out toggle buried deep in Instagram settings. This meant that account owners had to hunt down the setting themselves and turn it off. The union representing actors, SAG-AFTRA, quickly told members to opt out and published instructions to help them find the setting.

Notably, Meta did not notify people when their accounts had been used to create images. There was no alert, no audit trail visible to the person being pictured. The only way someone might learn their likeness had been used was to stumble across an AI-generated image of themselves elsewhere.

The Key Issue: Opt-Out vs. Opt-In

The core dispute is about how to handle permission for this kind of tool. Meta chose what's called an "opt-out" system: everyone's account was automatically available for the AI reference tool, unless they specifically went into settings and disabled it. The company placed the burden entirely on users to discover the feature existed and turn it off.

An alternative would be an "opt-in" system, where no one's account is used for image generation unless they explicitly agree to it first. Think of it like the difference between a newsletter that signs everyone up and makes them unsubscribe if they don't want it, versus a newsletter that only sends to people who actively sign up.

Meta's choice of opt-out meant that public visibility of a photo was treated as permission to use it. But critics argued — and the company eventually agreed — that public availability should not be the same as consent.

Why Meta Pulled It Entirely

By July 10, Meta disabled the feature outright rather than fixing it. The company could have redesigned the consent system, added notifications to people whose accounts were referenced, or limited the tool to only the account owner's own photos. Instead, it removed the feature entirely.

The most likely explanation is that Meta determined no quick fix would satisfy the legal risk or public reaction. Disabling the feature altogether was faster than rebuilding how it asks for permission. The company has not explained its reasoning.

The Broader Pattern

Meta has collided with likeness and consent issues multiple times while building AI image tools across its platforms. Each time, the company launches a broad capability, relies on opt-out settings as its permission framework, then retreats when public reaction is strong enough.

This recurring cycle points to a real question about how AI tools should be designed from the start. In my view, consent and permission should not be treated as an afterthought buried in a settings menu. They should be built into the tool's design from the beginning, before launch. A public photo being visible to everyone does not automatically mean it is licensed for a machine to remix it into new images of that person.

For regulators, unions, and everyday users, this episode sends a signal about where lines are being drawn around AI image generation in 2026: public availability alone is not enough to justify using someone's likeness without their knowledge, and the silent operation of the tool — no notification, no warning — made the situation worse, not better.

Meta has not said whether a revised version of the Muse Image reference feature will return.