Muse Image by Meta

Muse Image is Meta Superintelligence Labs' first in-house image model, announced on 7 July 2026 under the internal name Mango and rolled out free inside the Meta AI app and website, Instagram Stories and WhatsApp, starting in the United States. It sits alongside Muse Spark, the lab's language model, rather than on top of the older Llama image stack.

What makes it behave differently is that it plans before it draws. Given a prompt, the model works out the layout, looks up real-world details on the web when the request needs facts, and can fuse several reference images into one coherent result; edits stay iterative, so a second and third pass keep the image consistent instead of starting over. You can circle or annotate a region to show it what to change, and it renders legible in-image text - enough for step-by-step guides, infographics and working custom QR codes. A presets panel covers the frequent asks (restore an old photo, try a hairstyle, turn a selfie into a claymation or 16-bit character), and a room can be redesigned using real items listed on Facebook Marketplace, the integration Meta is building the feature toward.

Everyday creation is free, with heavier use behind Meta's paid subscription once you pass a threshold the company has not published, and outputs carry Meta's hidden Content Seal provenance marker. Two caveats before you lean on it: you can @-mention a public Instagram account and have its photos pulled into a generation - the setting can be switched off, and it drew criticism at launch - and availability still skews to the US. It is the sensible default when the finished image needs to live in a Story or a WhatsApp thread rather than a design file, and Meta previewed Muse Video from the same base model the day it shipped.