Meta ships Muse Image and previews Muse Video, its answer to Nano Banana and Sora

Meta Superintelligence Labs has shipped Muse Image, which it describes as its most advanced image generation model yet, and shared an early preview of Muse Video, according to Meta's own AI blog. Muse Image works differently from a typical image generator: it operates as an agent that can invoke tools to improve accuracy, including executing code to produce accurate plots and QR codes, searching the web to ground images in real, factual information, and self-refining its own outputs through internal reasoning before finalizing an image. It can also edit images with precision across multiple turns and compose new images from multiple reference photos. Muse Video, built on the same underlying foundation, adds native audio support and 'exceptional visual fidelity,' though it remains in preview rather than general release. Meta has built an invisible watermarking system called Content Seal directly into both models' outputs, designed to survive cropping, compression, resizing, or even screenshotting, so AI-generated images can still be verified as such after being shared or edited. As of July 5, Meta says Muse Image held the No. 2 spot on Arena's leaderboards for text-to-image generation, single-image editing, and multi-image editing, while Muse Video ranked No. 3 for text-to-video — putting Meta's media-generation stack in direct competition with Google's Nano Banana and OpenAI's Sora.