Technology

Meta Pulls Instagram AI Feature That Let Users Remix Others' Photos Without Notice

Martin HollowayPublished 4d ago4 min readBased on 3 sources
Reading level
Meta Pulls Instagram AI Feature That Let Users Remix Others' Photos Without Notice

Meta has removed an Instagram feature that let users generate AI images by @-mentioning public accounts as visual references, days after rolling it out TechCrunch.

The feature was part of Muse Image, an AI image generator built by Meta Superintelligence Labs. It shipped earlier this week and was withdrawn by Friday, July 10, 2026 TechCrunch. In practice, any user could tag a public Instagram account and have Muse Image produce a picture using that account's photos as a reference, with no notification sent to the person whose images were being drawn on TechCrunch.

Meta confirmed the rollback in a post on about.instagram.com, stating: "We've heard the feedback that this feature missed the mark, so it's no longer available" TechCrunch. The company did not elaborate on what specifically "missed the mark," and did not detail how many images were generated during the feature's brief availability.

Dylan Byers, founding partner of Puck News, was first to report the withdrawal, citing scrutiny from users and from talent agencies including Creative Artists Agency (CAA) TechCrunch. CAA's involvement points to a specific pressure point: agencies representing public figures and creators have a direct commercial interest in how their clients' likenesses get reused, and a no-notification AI remix tool touching any public account, celebrity or otherwise, sits squarely in that concern.

The timeline suggests the backlash was building before Meta's official statement. TechCrunch published a guide on July 9, one day ahead of the removal announcement, walking users through how to opt out of Muse Image using their photos TechCrunch. That opt-out setting was accessible via Instagram's profile settings, tucked behind the three-horizontal-lines menu TechCrunch. The existence of an opt-out at launch indicates Meta had anticipated some consent concerns, but the mechanism was opt-out rather than opt-in and buried deep enough in settings that most account holders would have had no practical way of knowing their photos were eligible for reference use at all.

That design choice is the crux of the story. An opt-out defaulting to "on" places the burden of discovery and action on the person whose content is being used, rather than on the party building the generative tool. It's a pattern familiar from data-scraping and model-training controversies across the industry — default-on data use, paired with an obscure toggle, tends to generate exactly this kind of reactive backlash once users notice after the fact rather than before.

Worth flagging: the speed of the reversal, days rather than weeks, suggests Meta's internal risk calculus shifted quickly once organized pushback from agencies materialized, rather than in response to individual user complaints alone. Consumer-facing generative AI features that touch third-party likeness or image data have repeatedly proven more legally and reputationally fraught than internal or first-party training use cases, and companies moving fast to ship them have just as often moved fast to retract them once professional stakeholders — legal teams, agencies, rights holders — get involved.

Meta has not said whether Muse Image itself, apart from the @-mention reference capability, remains available on Instagram, nor whether the company plans to reintroduce a revised version with clearer consent mechanics. The about.instagram.com post frames the removal as responsive to feedback rather than as an admission of a policy or design flaw, language that leaves open the possibility of a retooled relaunch.

In this author's view, the episode is a useful data point on how platforms are still calibrating consent norms for generative features that operate on other people's content rather than the user's own. Photo tagging, geotagging and reshare mechanics on Instagram have two decades of established norms behind them. Applying a public account's likeness as generative-AI training material in real time is a different category of use, and the instinct to default it to "on" reflects how unevenly that distinction is currently being applied across the industry, Meta included.