Meta’s Muse Image Can Put Your Public Photos in Someone Else’s AI Art. No Notification.

On July 7, 2026, Meta shipped Muse Image, the first image model out of its Superintelligence Labs, and buried inside the launch is a setting most creators have never seen: someone can tag your public Instagram profile, pull your face into an AI-generated image, and you will never be told it happened. Meta’s own terms spell it out. “People may be able to create content with your Instagram content using AI features at Meta,” the policy reads, and “you will not be notified about content created using AI features at Meta.”

That is the part of this launch that should change what you do this week. The technology underneath it is genuinely impressive, and I want to give it a fair hearing. But the default that ships alongside it is the kind of quiet operational decision I have spent twenty years flagging in IT and operations work, and it is worth understanding before you decide how much of Meta’s stack to build on.

What Muse Image actually is

Most image generators map a text prompt to pixels in one pass. Muse Image does something different, and the difference is the reason it matters. Meta built it to run as an agent. Instead of generating once and stopping, it “invokes search and coding tools to improve accuracy, self-refines its own generations, and improves through scaling test-time compute,” per Meta’s launch post.

Unpack that and three things stand out for anyone who makes visual content for a living.

First, it searches the web mid-generation. If you prompt something factual, a real product, a current event, a specific landmark, the model can pull real references to ground the output instead of hallucinating. Meta says this “improves factual accuracy on knowledge-intensive prompts, particularly those involving current events and real-world facts.” For a creator who has ever fought an image model over a logo that came out wrong or a piece of text that came out as gibberish, that is a meaningful shift.

Second, it writes and runs code. Meta’s engineers found the model learned to “write and execute code that produces accurate plots and QR codes.” A working QR code out of an image generator sounds like a party trick until you are a creator who needs a scannable link on a thumbnail or a merch mockup, and every other tool renders a decorative square that scans as nothing.

Third, it thinks longer when you let it. Quality scales “approximately log-linear” with the compute you give it. The model reasons more, calls more tools, and runs more self-refinement passes. Meta describes the self-refinement as something that “emerged organically during training”: the model reflects on a draft and decides whether to make a local edit, regenerate from scratch, or reach for a tool. That is closer to how a working designer iterates than how a one-shot generator behaves.

The editing side is strong too. Muse Image composes from multiple reference images, holds coherence across editing turns, and lets you interleave text and images inline in a prompt. On the LMArena leaderboards dated July 5, it landed at #2 for text-to-image, single-image editing, and multi-image editing. Meta also previewed Muse Video, built on the same base with native audio, sitting at #3 for text-to-video and “coming soon to creators.”

None of this is vaporware. It is live right now in the Meta AI app, at meta.ai, in Instagram Stories in the US, and in WhatsApp in a limited set of countries, with Facebook coming soon.

The consent default is the story

Here is where my operations instinct kicks in. When you evaluate any vendor, the feature list is the marketing. The defaults are the product. And Muse Image ships with a default that treats other people’s likenesses as raw material.

The mechanism: you can tag a person with a public profile, and the model will use their public images to generate new content featuring them. The person is never notified. Meta says users retain control through settings, but this runs opt-out, not opt-in. You are enrolled by default and have to go find the toggle to leave.

Read that as a creator, not as a casual user. Your face is your brand. If you post publicly, and most of us have to, your likeness is now a default input to a generation model that anyone can invoke. The pushback was immediate. One widely shared post on X called it “a privacy landmine waiting to detonate,” and the phrase fits. This is not a hypothetical about some future misuse. It is the shipped behavior on day one.

I have sat in enough vendor reviews to know the pattern. A company under pressure to justify AI spending picks the default that maximizes data flow and shifts the burden of objection onto the user. Meta is under exactly that pressure right now, which is the context for the whole Muse rollout.

Follow the money and the defaults make sense

Muse Image is free for everyday creation, but the ceiling is low. Power users and creators who want volume or advanced features have to pay into Meta’s subscription tiers, which the company began testing in late May. Meta One Plus runs $7.99 a month, Meta One Premium runs $19.99, and the creator and business tiers climb to $14.99 and $49.99, with the top tier finally offering human support for Instagram and Facebook pages.

So the picture is: a model that trains and generates on a firehose of public user content, offered free at the shallow end to drive adoption, with a paywall for anyone who actually depends on it, all launched in the same quarter Meta needed to show investors it could turn AI spending into revenue. The opt-out default is not an accident. It is the design that makes the economics work.

That does not make the tool unusable. It makes it a tool you use with your eyes open, the way you would treat any platform that monetizes the data you hand it.

What creators should actually do

I am not going to tell you to boycott Meta Muse Image. That is not practical advice for anyone whose audience lives on Instagram, and the model is good enough that ignoring it would be a mistake. Here is the more useful version.

Go into your Meta AI settings and find the control that governs whether others can use your content in AI features. Turn it off if you do not want your likeness pulled into strangers’ generations. Do this even if you never plan to touch Muse Image yourself, because the setting is about what others can do with you, not what you do with the tool.

Treat anything you generate as watermarked, because it is. Meta embeds what it calls Content Seal, an invisible provenance signal that “stays intact, even when cropped, compressed, resized, or screenshotted,” with a detector at meta.ai/identification. That is genuinely good for provenance and disclosure, and it is also a reminder that Meta knows exactly what came out of its model. Plan your disclosures accordingly, especially given the tightening platform rules on labeling AI content.

Do not make Muse Image your only image pipeline. The agentic search and code features are real advantages, but vendor concentration is a risk I flag on every operations engagement. Keep a second tool in rotation so a policy change or a pricing move does not strand your workflow. If you want a map of the alternatives, our 2026 rundown of AI image tools for creators covers where each one is strong, and the Nano Banana workflow and ChatGPT Images 2.0 both hold up as everyday drivers.

If you generate images of real people, including yourself, understand the rights you are standing on. The consent question Muse Image raises is the same one the courts are still sorting out, and our breakdown of AI image copyright and likeness law in 2026 is the primer I would read before building a business on any of this.

The honest verdict

Muse Image is the best argument yet that agentic generation, a model that searches, codes, and self-corrects, is where image tools are heading. On capability, Meta earned its ranking. If you shoot product photos, build thumbnails, or need factual accuracy in your images, it deserves a spot in your testing this month.

But it arrives with a consent model that quietly conscripts every public profile, yours included, into its training and generation loop, and asks you to opt out rather than in. That is not a reason to avoid it. It is a reason to configure it before you use it, keep a backup tool, and remember that on Meta’s platforms the default always favors Meta. This is the same lesson every subscription rollout teaches, which is why it is worth reading alongside what Meta One actually charges creators. The tool is powerful. The terms are the price. Read them.

Sources: Meta AI: Introducing Muse Image and Muse Video, Meta Newsroom, TechCrunch, Axios, CNBC on Meta One pricing.