Meta Pulled an Instagram AI Feature Within Days. Here's Why.

Meta removed an Instagram feature that let users generate AI images by tagging other people's public accounts as visual references, just three days after launching it TechCrunch. The feature was part of Muse Image, an AI image generator built by Meta's Superintelligence Labs division. It shipped early in the week and was gone by Friday, July 10, 2026.
The way it worked was straightforward but troubling. Any Instagram user could mention a public account using an @ tag, and Muse Image would generate a picture using that account's existing photos as a style or composition reference. The person being tagged received no notification TechCrunch. Meta confirmed the removal in a post on about.instagram.com: "We've heard the feedback that this feature missed the mark, so it's no longer available" TechCrunch. The company offered no specifics on what "missed the mark" meant or how many images were generated during the brief rollout.
Dylan Byers at Puck News broke the story of the withdrawal and reported that pushback came from both ordinary users and major talent agencies, including Creative Artists Agency TechCrunch. That detail matters: agencies representing celebrities and creators have a direct financial stake in how their clients' images are used and remixed. A tool that lets anyone generate AI pictures based on a famous person's face, without their knowledge, sits directly in that area of concern.
The backlash timeline tells part of the story. TechCrunch published instructions on July 9 for how users could opt out of having their photos used by Muse Image — a full day before Meta's removal announcement TechCrunch. The opt-out switch was buried in Instagram's settings menu, behind the three-horizontal-lines icon TechCrunch. The fact that an opt-out existed at launch shows Meta anticipated some concerns, but it was opt-out rather than opt-in — meaning the feature was on by default, and most account holders would have had no idea their photos were being used as reference material.
That design choice is the core issue. When a feature defaults to "yes" and the "no" button is hidden in a menu most people never visit, the burden falls on the person whose content is being used to figure out what's happening and turn it off. This is a pattern we have seen across the industry repeatedly. Data scraping, model training, and other uses of people's information tend to follow the same structure: on by default, opt-out somewhere in settings. The result is backlash once people find out after the fact rather than before.
The speed of Meta's reversal matters. The company pulled the feature within days, not weeks — a shift that seemed to happen once organized pressure from talent agencies arrived, rather than from individual user complaints alone TechCrunch. We have seen this pattern before. Consumer-facing AI tools that touch other people's faces, names, or images have proven more legally risky and reputationally costly than AI systems that use training data behind the scenes. Companies that move fast to launch them often move just as fast to pull them back once lawyers and professional representatives get involved.
Meta has not said whether Muse Image itself — the core AI image generator — remains available on Instagram, or whether the @-mention reference feature will return in a different form with clearer permission rules. The company's statement frames the removal as a response to feedback rather than an admission of a design mistake, which leaves the door open for a revised version to reappear later.
The broader context here is that social platforms are still learning how to handle permission and notification for AI features that work with other people's content, not your own. Tagging friends, sharing location, resharing photos — these have two decades of norms behind them on Instagram and other platforms. Using an AI tool to remix someone else's face or image style in real time is a newer category, and the choice to turn it on by default reflects how unevenly the tech industry is still thinking about consent in these cases. Meta included.


