Meta has released Muse Image, the first AI media generation model from its Superintelligence Labs division, now powering image tools across the Meta AI app, Instagram, and WhatsApp. The company also previewed Muse Video, a forthcoming video generation model, as part of a broader push to replace its Llama lineup with the Muse family.
Alexandr Wang, who Meta hired to lead Superintelligence Labs last year, described Muse Image as "agentic" on Threads, explaining it works with the Muse Spark large language model to reason through prompts, search the web, and plan before generating. This marks a strategic shift from Meta's previous Llama-based approach toward more integrated, reasoning-driven AI tools.
Muse Image is now live on iOS and Android via the Meta AI app, with Facebook and Messenger integration expected soon. The model enables users to create and edit AI-generated images directly within social feeds, including options to incorporate other Instagram users into generated photos, though specific output guardrails remain unclear.
The launch signals Meta's intensifying competition in generative AI, positioning its proprietary models against offerings from OpenAI, Google, and others. By embedding creation tools directly into its social ecosystem, the company risks amplifying concerns over synthetic content provenance and user consent, particularly around unauthorized likeness use.
Critics argue that agentic AI models, which search the web and plan autonomously, may introduce unpredictable outputs or privacy risks. Meta has not detailed how Muse Image handles sensitive prompts or prevents misuse, nor disclosed the training data composition for its new media models.