Meta’s Muse Image puts tool-using generation inside Instagram and WhatsApp

Image generation is moving from standalone demos into the places where people already post. Meta’s Muse Image release makes the model available in the Meta AI app, on meta.ai, in Instagram Stories in the U.S., and in WhatsApp in limited countries, with Facebook support planned later.

The technical bet is that an image model should behave more like an agent than a direct prompt-to-pixel system. Meta says Muse Image can use web search to ground visually specific prompts in current facts and references, write and execute code for plots and QR codes, and revise its own outputs when details are wrong. It also connects with Muse Spark so the systems can share tools and planning for broader media workflows.

Meta published concrete ranking claims. As of July 5, 2026, Muse Image held the No. 2 position on Arena by human-preference Elo for text-to-image generation, single-image editing, and multi-image editing. Muse Video, built on the same pretraining base, is still an early preview, but Meta says it ranks No. 3 on Arena for text-to-video. The company also named current gaps for video, including audio-video synchronization and physically accurate fast motion.

The distribution surface makes the release more consequential than another leaderboard update. Instagram and WhatsApp give Muse Image a path into everyday visual communication, creator workflows, and small-business marketing assets. That also raises the burden for provenance. Meta is shipping Muse Image with Content Seal, an invisible watermark intended to survive cropping, compression, resizing, and screenshots, and it is previewing a detection tool for checking whether an image carries the signal. The next test is whether quality, controls, and labeling hold up at social-network scale.