Meta on Tuesday launched Muse Image, its first fully in-house AI image generation model, as the company moves away from relying on third-party models such as Midjourney and Black Forest Labs. Developed by Meta Superintelligence Labs under Chief AI Officer Alexandr Wang, the model is rolling out for free through the Meta AI app, Instagram Stories in the US, and WhatsApp in select markets.
Internally codenamed Mango, Muse Image is the second major product from Meta's new AI division after April's Muse Spark language model. Unlike conventional text-to-image systems, Muse Image is designed to function as an AI agent, capable of reasoning through prompts, searching the web for real-time information, generating code for tasks such as QR codes or charts, and refining its own outputs before presenting results.
Meta claims the model performs better than Google's Nano Banana 2 on multi-image editing tasks but trails OpenAI's GPT Image 2, based on the company's internal benchmarks. Alongside Muse Image, Meta also previewed Muse Video, an unreleased AI video generation model, and introduced Content Seal, an invisible watermark designed to help identify AI-generated images after editing or compression.
The launch quickly drew criticism over a feature that allows Meta AI users to tag any public Instagram account and use that person's publicly available photos to generate AI images. The feature is enabled by default, with users required to opt out if they do not want their public images used this way.
Meta says users remain in control through privacy settings, but critics have compared the feature to previous privacy controversies involving the company, including Facebook's facial recognition system, which was shut down in 2021 following regulatory scrutiny and legal challenges.
The concerns also revive broader questions around Meta's handling of user data, years after the company paid a record $5 billion FTC fine over the Cambridge Analytica scandal.
