Meta launched its first image generation model built by Meta Superintelligence Labs on July 7, 2026, then withdrew one of the tool's launch features three days later after users objected to how it handled public Instagram photos. The model, called Muse Image, rolled out simultaneously inside the Meta AI app, on Instagram Stories, and in WhatsApp chat, according to the company's own announcement published on its Newsroom site. Meta has not yet opened the tool to advertisers directly, though it says that access is coming through Advantage+ creative within weeks.

The launch and the reversal happened inside the same official post. Meta did not publish two separate announcements; instead, the company appended an update to its original July 7 text at 3:45 p.m. Pacific Time on July 10, confirming that a feature letting people @-mention public Instagram accounts to pull those accounts' photos into new AI-generated images was "no longer available." Meta wrote that its intent had been to give people a useful creative tool while letting them control whether their own public content could be referenced this way, and that it had heard feedback the feature "missed the mark."

What Muse Image actually does

According to Meta, Muse Image is designed to understand complex, conversational prompts and to blend multiple photos together into a single finished image, rather than requiring a person to specify technical parameters or use a separate editing application. A person can type a request in plain language, reference an existing photo, and receive a result they can download or post directly to a feed, story, or chat thread. Meta describes the underlying capability as advanced reasoning, though the company's announcement does not define that phrase technically or compare it against a named competing model.

The company's own demonstration prompt, included in the announcement, illustrates the level of descriptive detail the model is built to parse: a street-photography scene of a man walking through a crowded summer block party in Brooklyn, carrying barbecue in one hand and lemonade in the other, shot with a shallow depth of field and natural film grain. Whether that single example generalizes to the broad range of prompts everyday users will submit is not something Meta's announcement addresses; it is, after all, the company's own chosen showcase.

Three specific interaction methods sit alongside the core generation capability. Presets, described as suggested prompts, are meant to give people a starting point when they are not sure how to phrase a request. The @-mention function, separate from the withdrawn Instagram-tagging feature, lets someone add their own uploaded photos into a generation session so the model can incorporate a specific person, pet, or object into a new image. A markup tool lets a person tap directly on a generated image to circle, sketch, or annotate the exact area they want changed, and Meta says the assistant retains the surrounding conversational context so that follow-up edits do not require restarting the process from a blank prompt.

Text rendering and practical use cases

One detail in Meta's announcement is worth flagging technically: the company states that Muse Image renders text cleanly inside generated visuals, which it frames as enabling requests such as a how-to guide or an infographic on a specific subject where the on-image text needs to remain legible and styled consistently. Text rendering has historically been a weak point for image generation models generally, producing garbled or nonsensical characters where legible words were requested; Meta's claim, if accurate across a broad range of use cases, would represent a meaningful practical improvement for anyone using the tool to produce instructional or informational graphics rather than purely artistic images. Meta's announcement does not include comparative benchmarks against other image models on this specific capability.

The three-day retraction

Muse Image shipped with a capability that let a user @-mention any public Instagram account, not just their own uploaded photos, and pull that account's existing pictures into a new AI-generated image. Meta's own updated text describes the intent behind the feature in fairly plain terms: the company wanted to give people a useful creative tool while also letting them control whether their public content could be referenced in this way. The two halves of that sentence turned out to be in tension. A public Instagram account, by definition, belongs to somebody who has not necessarily consented to having their photos become raw material for someone else's generated image, and the feature offered no separate opt-in step for the account being referenced.

Meta's language in the July 10 update is notably direct for a corporate product statement. The company said plainly that the feature "missed the mark" and that it was "no longer available," without elaborating on how many images had been generated through the feature during the roughly three days it was live, and without disclosing how many Instagram accounts had their photos referenced without the account holder's involvement. Meta drops Instagram tagging tool as 8 of 10 apps default users in documented the retraction in detail, including on-the-record comments from Surfshark's Luís Costa connecting the episode to a broader industry pattern of AI systems defaulting users into data collection rather than asking first.

The sequencing is the part worth sitting with. A feature can move from public launch to public withdrawal in under a week when the objection is sharp enough and visible enough, and Meta's own choice to fold the retraction into the same Newsroom post, rather than issuing a fresh statement, suggests a company treating the correction as a continuation of one product story rather than as a separate incident requiring its own explanation.

Where the model shows up across Meta's apps

Muse Image is not confined to a single surface. Meta says the model now powers more than 30 new AI-powered effects for Instagram Stories, giving people a substantially larger set of stylistic transformations to apply to their own content than existed before the launch. On WhatsApp, people can generate images directly inside their chat conversations with Meta AI, though Meta specified that this capability is starting in a limited set of countries, with additional locations expected to follow; the company's announcement does not name which countries were included at launch.

Meta says it will bring Muse Image to more countries generally in the near term, and separately extend the model to Facebook and Messenger, along with additional surfaces on Instagram and WhatsApp beyond what shipped on July 7. No specific date accompanies either expansion. Using Meta AI with Muse Image is free for what the company calls everyday creation, while people who want higher usage volumes can access it through Meta's existing subscription plans; the announcement does not specify a usage threshold that separates free access from the point at which a subscription becomes necessary.

Advertiser access is still weeks away

For the marketing and advertising community, the detail carrying the most direct operational weight is also the one Meta describes in the vaguest terms: in the company's own words, advertisers and agencies will be able to tap into Muse Image through Advantage+ creative in the coming weeks. That places Muse Image's advertiser-facing debut at some unspecified point after July 7, 2026, rather than concurrent with the consumer launch, and it means that, as of this announcement, no advertiser has yet used the model to produce paid creative through Meta's ad tools.

Advantage+ creative is the branding under which Meta packages its automated ad-creative features, and it already carries substantial adoption momentum. Meta Q1 2026: $56.3B revenue as AI tools double advertiser adoption reported that more than 8 million advertisers were using at least one of Meta's AI ad creative tools by the end of the first quarter of 2026, doubling from 4 million roughly four months earlier, with small and medium-sized businesses accounting for the large majority of that growth. Muse Image, once it reaches Advantage+ creative, would extend that same toolkit with a newer underlying model, though Meta's Newsroom post does not describe what, if anything, changes technically between Muse Image inside Meta AI and Muse Image inside the advertising product.

That distinction between a consumer launch and an advertising launch matters because Meta's recent history with automated creative tools inside Advantage+ has not been uniformly smooth. Meta auto-enrolled REI into AI: what Cannes week revealed about brand control detailed how, in late June 2026, the outdoor retailer REI discovered that Meta had automatically enrolled its advertising account into a separate AI image-modification tool without the retailer's knowledge, a setting that subsequently altered a vendor-supplied bicycle photograph and produced a visibly deformed image that ran on Instagram for roughly a week before REI's marketing team noticed and had it removed. That earlier episode did not involve Muse Image specifically, since it predates this model's launch by several weeks, but it does establish a pattern that advertisers evaluating Meta's forthcoming Advantage+ integration will likely want to weigh: default settings inside Meta's ad tools have, at least once recently, applied AI modification without an explicit advertiser opt-in.

Whether Muse Image's rollout into Advantage+ creative will require advertisers to actively opt in, or whether it will arrive enabled by default the way some earlier Advantage+ features have, is not addressed in Meta's July 7 announcement. Advertisers managing accounts at scale, particularly through agency relationships where day-to-day settings are not reviewed line by line, may find that question worth raising directly with their Meta account representatives once the advertiser-facing rollout actually begins.

Muse Image sits inside a larger model family

Muse Image did not appear in isolation. It is the second named product to emerge from Meta Superintelligence Labs following Muse Spark, which Meta introduced on April 8, 2026 and described at the time as the division's first model, purpose-built to prioritize people. Muse Spark was positioned primarily as a smarter conversational assistant; Muse Image, according to Meta's framing in this announcement, functions instead as what the company calls a creative partner that understands a user's world, translating typed ideas directly into finished visuals.

Two days after Muse Image's launch, on July 9, 2026, Meta released a distinct follow-on model, Muse Spark 1.1, opening it to outside developers through a new public preview of the Meta Model API. Meta's Muse Spark 1.1 gains 1 million token context for developers today covered that release separately, noting that Muse Spark 1.1 is a multimodal reasoning system aimed at agentic and coding tasks rather than image generation, and that Meta's own announcement of it makes no reference to Ads Manager, campaign automation, or advertising measurement infrastructure. The two releases, arriving within 48 hours of each other, are still built by the same research organization, established as Meta Superintelligence Labs on July 30, 2025, and Meta's own language frames both as steps toward what it calls personal superintelligence, a long-term company vision rather than a specific claim about either individual model.

Meta's announcement closes by gesturing toward a third product still in progress. The company states that images are just the beginning, and that Muse Video is already in development, without offering a timeline, a description of its planned capabilities, or any indication of which of Meta's apps it would eventually reach. Whether Muse Video follows the same launch-then-adjust pattern that Muse Image exhibited in its first three days is, naturally, not something the current announcement can answer.

Why the sequencing matters to marketers

For anyone planning creative production or campaign strategy around Meta's platforms, three separate facts sit alongside each other in this announcement, and each carries a different practical implication. First, a genuinely new image-generation capability now exists inside apps that reach billions of people, with in-image text rendering and multi-photo blending as its most technically distinctive features. Second, that same capability shipped with at least one feature significant enough to draw public objection and prompt a reversal within three days, which suggests Meta's internal review process for consumer-facing AI features did not fully anticipate how an @-mention-a-public-account function would be received once it was live. Third, the advertising-facing version of the tool has not shipped yet, which gives marketing teams a window, however brief, to think through governance questions before Advantage+ creative access arrives, rather than discovering settings after the fact the way REI did with a different Meta AI tool weeks earlier.

None of this suggests Muse Image is unusual by the standards of how quickly generative AI features have moved from launch to iteration across the industry over the past two years. What is unusual, and worth marketers' attention specifically, is Meta's own choice to narrate both the launch and the walk-back inside a single running document rather than treating them as separate news events, and its explicit statement that agencies will get access to a tool whose consumer version needed a correction within its first week.

Timeline

  • July 30, 2025: Meta establishes Meta Superintelligence Labs, the research organization that later builds both Muse Spark and Muse Image.
  • April 8, 2026: Meta introduces Muse Spark, describing it as Meta Superintelligence Labs' first model, purpose-built to prioritize people.
  • July 7, 2026: Meta launches Muse Image across the Meta AI app, Instagram Stories, and WhatsApp chat, including a feature letting users @-mention public Instagram accounts to reference those accounts' photos.
  • July 9, 2026: Meta releases the separate Muse Spark 1.1 model and opens a public preview of the Meta Model API to outside developers.
  • July 10, 2026, 3:45 p.m. Pacific Time: Meta updates its Muse Image announcement, confirming the Instagram @-mention feature is no longer available following user objections.
  • Ongoing at time of writing: Meta states that advertisers and agencies will gain access to Muse Image through Advantage+ creative in the coming weeks, with no confirmed date.

Summary

Who: Meta, through its Meta Superintelligence Labs research division, built and launched Muse Image. The retracted feature affected any Instagram user with a public account, whether or not they used Meta AI themselves.

What: Meta introduced its first image generation model from Meta Superintelligence Labs, capable of blending multiple photos, editing images through on-screen markup, and rendering legible text inside generated visuals. Three days after launch, Meta withdrew a feature that let users @-mention public Instagram accounts to reference those accounts' photos without the account holder's involvement.

When: Muse Image launched July 7, 2026. Meta withdrew the Instagram @-mention feature on July 10, 2026, at 3:45 p.m. Pacific Time. Advertiser access through Advantage+ creative is described only as coming in the weeks following the announcement.

Where: The launch spans Meta's global product footprint through the Meta AI app, Instagram Stories, and WhatsApp chat, with WhatsApp image generation starting in a limited, unnamed set of countries.

Why: The announcement matters to marketers because it previews a new AI creative tool headed for Meta's advertising infrastructure while simultaneously demonstrating, through the three-day retraction, that Meta's own review process for consumer AI features did not anticipate the public reaction to at least one launch capability, a pattern advertisers may want to weigh before Advantage+ creative access begins.