Meta has rolled back a controversial feature in its new Muse Image model, which allowed generating images using public Instagram profiles as prompts. The company quickly responded to the wave of criticism and removed this option just three days after its launch.
On July 7, Muse Image was introduced as the first image generation model from Meta Superintelligence Labs, integrated into Meta AI. A key feature was the ability to specify a public Instagram account in the prompt — the neural network automatically pulled open photos of the user to create new content. However, by July 10, Meta released an update acknowledging that the feature "did not meet user expectations regarding privacy."
The problem was that the option was enabled by default, creating serious privacy risks. Actress Hannah Einbinder publicly called for disabling this feature, and the SAG-AFTRA union deemed the situation unacceptable. According to their statement, any mechanism other than explicit and conspicuous consent for using images is a "complete miscalculation of public opinion" and carries obvious dangers and harm.
After the feature was disabled, the union supported Meta's decision, emphasizing that using public photos without direct permission was a grave mistake. It is important to note that the Muse Image model itself as a product remains available in Meta AI for generating and editing images based on text prompts and uploaded photos. Additionally, Muse Image is already used in over 30 new AI effects for Instagram Stories and in WhatsApp chats with Meta AI in a limited number of countries.
As a reminder, on June 9, Meta also introduced the multimodal model Muse Spark 1.1, focused on agent tasks and coding. However, this incident with Muse Image shows that even industry giants can make serious mistakes in assessing privacy boundaries, especially when it comes to generative AI working with real user data.
My expert conclusion: This case is a vivid example of how the pursuit of innovation can clash with strict privacy requirements. Meta clearly underestimated the community's reaction, and although the feature rollback was swift, reputational damage has already been done. In the era of AI regulation, such miscalculations are becoming increasingly risky — users and regulators will not forgive the use of personal data without explicit consent.