Meta launched Muse Image on Tuesday, the first image generation model built by its Superintelligence Labs division, and immediately wired it into Meta AI, Instagram and WhatsApp, with Facebook and Messenger to follow. The model replaces Llama-based image tooling across those surfaces and will power 30 new AI effects rolling out to Instagram Stories in the US before expanding to other countries. Its most consequential feature is social: users can @-mention another Instagram account inside a prompt and have that person's likeness rendered into the output.
Muse Image is part of a new Muse family of models that supersedes Meta's Llama lineup for generative media. It is paired with Muse Spark, a language model that Meta describes as the reasoning layer of the stack. Alexandr Wang, hired last year to lead Superintelligence Labs, called the combined system agentic on Threads, saying Muse Spark reasons through prompts, searches the web and plans before Muse Image generates a frame.
That architecture is the pitch: Meta is not shipping a single-shot text-to-image endpoint but a planner-plus-generator pipeline aimed at the kinds of composite requests social users actually make — invitations, room redesigns, group scenes with real people in them. The @-mention feature is the sharpest example. Meta says tagging a username lets Meta AI use public photos to build a visual, effectively turning any public Instagram profile into a reference input for someone else's generation.
“to reason through your prompt, search the web, and plan before it generates.”— Alexandr Wang, Head of Meta Superintelligence Labs
Key facts
- 01Muse Image is the first image generation model from Meta's Superintelligence Labs, replacing Llama-based tooling across Meta AI, Instagram and WhatsApp.
- 02Users can @-mention Instagram accounts inside prompts, letting Meta AI incorporate a tagged user's likeness from their public photos.
- 03The model will power 30 new AI effects rolling out to Instagram Stories in the US first, then other regions.
- 04A companion Muse Video model was teased by Superintelligence Labs head Alexandr Wang, framed as competitive on prompt adherence and temporal consistency.
- 05Muse Image is paired with Muse Spark, a language model that plans and web-searches before generation, making the pipeline agentic.
Meta says users can control how their content is reused for AI, but did not detail the granularity of those controls in Tuesday's announcement, nor whether the default is opt-in or opt-out. The company also did not disclose whether tagged users are notified when their likeness appears in another account's generation, or whether generated images carry provenance metadata.
The product surface goes well beyond portraits. Muse Image can redesign rooms using a source image pulled from Facebook Marketplace or elsewhere on the web, transform photos through suggested prompts, and let users draw directly on an image to steer edits before sharing to feed, story or chat. Design templates for postcards and invitations are built in. The 30 new Instagram Stories effects are the most visible consumer wedge, sitting inside a camera that already sees billions of daily uses.
Wang also teased a forthcoming Muse Video model on Threads, positioning it against the current leaders in text-to-video. He did not share benchmark figures, release timing or pricing, and Meta has not said whether Muse Video will ship inside Reels, the standalone Meta AI app, or both.
“competitive on prompt adherence, visual fidelity, temporal consistency.”— Alexandr Wang, Head of Meta Superintelligence Labs
The move consolidates Meta's generative stack under Superintelligence Labs, the unit Wang joined last year after Meta acquired his prior company's talent and staked the lab as the center of gravity for its frontier work. Retiring the Llama brand from consumer image tools is a notable internal shift: Llama remains Meta's open-weights text model line, but image and video are now Muse. It clarifies the org chart and, more importantly, gives Wang's team direct ownership of the outputs users see inside Instagram and WhatsApp.
The likeness feature is where the product will draw the most scrutiny. Letting one user drop another user into an AI-generated scene using only a public handle collapses the distance between reference art and impersonation, and puts pressure on Meta's moderation stack to catch harassment, non-consensual imagery and deceptive edits at scale. Meta's stated position — that public photos are fair game unless the user has adjusted controls — leans on the assumption that most Instagram users know those controls exist. Discord's recent moderation-model misfire, which wrongfully banned 8,000 users, is a useful reminder of how quickly automated enforcement at social-network scale can generate its own failure modes.
Muse Image lands Meta squarely in the middle of the AI-native social experience race, and the @-mention primitive is a genuinely new interaction pattern — the first time a major platform has made another user's likeness a first-class prompt token. If it holds up in practice, it converts Instagram's social graph into training-signal-adjacent generation fuel and gives Meta a distribution edge no standalone image generator can match. If the abuse surface proves as wide as it looks, the controls will need to be tightened publicly and quickly, because 30 new Stories effects times a billion accounts is a very short runway to get moderation right.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
