Meta ships Muse Image, an agentic image model ranking #2 on Arena

Meta Superintelligence Labs released Muse Image, which generates images as an agent — invoking search and code tools, self-refining its drafts, and scaling test-time compute like a language model — ranking #2 on Arena for text-to-image and editing, alongside an early-preview Muse Video (#3 for text-to-video, with native audio). The signal is architectural: image generation now rides the same reasoning-and-tool-use scaling curve as LLMs, and both models are live across the Meta AI app, meta.ai, and Instagram Stories in the US.

Source: ai.meta.com ↗