Meta's Superintelligence Labs has released Muse Image, its first AI image generation model, rolling it out across the Meta AI app, Instagram, and WhatsApp, with Facebook and Messenger support coming soon The Verge. The model belongs to the Muse family, which Meta is positioning to replace its Llama lineup The Verge.

Muse Image is billed by Meta as "agentic." Alexandr Wang, who leads Superintelligence Labs, said on Threads that the model works in tandem with Muse Spark — Meta's large language model — to reason through a prompt, search the web, and plan before actually generating an image, rather than mapping text to pixels in a single pass The Verge Threads. That framing distinguishes it from the prompt-in, image-out diffusion tools that have defined the category since Stable Diffusion and DALL-E popularized it.

The consumer-facing features lean hard into Meta's social graph, which is the part of this release worth the most scrutiny. Users can @ mention an Instagram account inside a Muse Image prompt, and the model will incorporate that account's likeness into the generated image; Meta says tagging a username lets Meta AI draw on that account's public photos to build the visual The Verge. Anyone who doesn't want their public photos used this way can adjust the setting through Instagram's help center controls Instagram Help. Beyond likeness-based generation, the model supports image transformation via suggested prompts, generation of designs such as invitations and postcards, room redesigns sourced from an image pulled off Facebook Marketplace or elsewhere on the web, and freehand drawing directly on a photo before sharing the edited result to an Instagram feed, story, or chat Meta. Instagram Stories will also get 30 new AI effects powered by Muse Image, launching first in the US ahead of other markets The Verge Instagram.

Meta has also confirmed a Muse Video model is planned, though no release date has been specified Meta AI Blog.

Muse Image is the second public release from Superintelligence Labs, following Muse Spark, which Meta introduced on April 8, 2026 as the first model out of the division and the engine now running the Meta AI assistant across WhatsApp, Instagram, Facebook, and Messenger Meta The New York Times. That launch came roughly three months after Meta closed the $14 billion arrangement that brought Wang in to run its AI efforts CNBC, one of the largest single talent-and-equity deals in the industry's history and a figure that set the bar against which everything Superintelligence Labs ships will be measured.

The developer-facing side of Muse Spark has not moved at the same pace as the consumer rollout. As of June 3-4, 2026, Meta had repeatedly delayed the Muse Spark API's release to third-party developers and had no scheduled date for it Reuters. Meta has not said whether Muse Image's underlying weights or API access will follow a different timeline, and the July rollout notice does not address developer access at all Meta.

There is also a regulatory thread running underneath this. On June 23, 2026, it was reported that the Trump administration is pressing Meta to submit its AI models for voluntary government review amid rising security concerns Reuters. Meta has not indicated whether Muse Image or Muse Spark would be covered by any such review, or how that pressure interacts with the API delays already reported.

The likeness-tagging feature is the part of this release that will draw the most attention from privacy-minded users and regulators alike, since it operationalizes a photo library that was built for social sharing into training material for on-demand generative output. Meta's answer — an opt-out buried in help center settings rather than an opt-in gate — follows a pattern familiar from Meta's prior data-use controversies, where the default favors platform utility and the burden of restriction falls on the user. Whether that default holds up under the kind of scrutiny already directed at the administration's review request is a separate question from whether the technology works, and it's one Meta has not yet had to answer publicly at scale.

Set against the API delays, the picture that emerges is a company moving fast on consumer features it controls end-to-end while proceeding more cautiously — or simply more slowly — on the parts of the stack it would have to hand to outside developers. That gap between what ships to a billion-plus users overnight and what ships to a sandbox of registered developers is not unusual for a platform company managing risk, but it is notable given how aggressively Meta priced its return to frontier AI competition with the Wang deal. The near-term test will be whether Muse Video arrives with the same consumer-first sequencing, and whether the agentic reasoning Wang described for Muse Image — web search and planning ahead of generation — proves durable once it is exposed to the volume and adversarial creativity of Instagram's user base.