Meta has shut down image generation from public Instagram photos: privacy proved stronger than AI ambitions
Meta has decided to disable a feature in its Muse Image model that allowed generating images based on mentions of public Instagram accounts. The corresponding update was made to the official announcement of the model just three days after its launch.
Controversial Option: From Excitement to Disablement in 72 Hours
On July 7, I analyzed in detail the launch of Muse Image — the first image generation model from Meta Superintelligence Labs, integrated into Meta AI. The key "feature" was the ability to add any public Instagram profile to the prompt so that the algorithm would use open photos to create new content. However, by July 10, the company announced the removal of this feature.
The official reason is that the option "did not meet user expectations in terms of privacy." In my opinion, this is a soft formulation for a serious failure in assessing public sentiment. The problem lay in the automatic activation of the feature: users did not need to give explicit consent for their public photos to be used for AI image generation.
Industry Reaction: A Critical Signal for All AI Products
Hannah Einbinder, a well-known actress, publicly called for the option to be disabled. The SAG-AFTRA union issued an even harsher statement, calling any mechanism other than explicit and noticeable consent a "complete miscalculation in assessing public opinion" and an "unacceptable risk." The organization supported Meta's decision to disable the feature but emphasized that the initial implementation was flawed.
What Remains: Muse Image Lives On, But Without the Instagram Dataset
It is important to understand: the Muse Image model itself as a product has not disappeared. It continues to operate in Meta AI for generating and editing images based on text prompts and uploaded photos. Moreover, Meta is already using the technology in 30+ new AI effects for Instagram Stories and in WhatsApp chats with Meta AI in several countries. As a reminder, on June 9, the company also introduced the multimodal model Muse Spark 1.1 for agent tasks and coding.
My analysis: This incident is a vivid example of how even giants like Meta can underestimate the "privacy factor" when implementing generative AI features. In the pursuit of innovation, the company forgot the basic principle: consent must be explicit, not implied. For the market, this is a signal: regulators and the public will harshly crack down on any attempts to use user data for AI training without transparent consent. Meta quickly corrected the mistake, but the reputational residue will remain.