Meta launches AI image generator Muse Image, raising privacy concerns

Meta promotes the product with the slogan that Muse Image is a model that "knows your world." Specifically, the company describes the tool as a "creative partner who knows your world and makes it easy to turn your ideas into high-quality images that you can download and share anywhere – including directly in your feed, your story, or your chat." The original German quote reads: "Muse Image ist Ihr kreativer Partner, der Ihre Welt kennt und es Ihnen leicht macht, Ihre Ideen in hochwertige Bilder umzusetzen, die Sie herunterladen und überall teilen können – auch direkt in Ihrem Feed, Ihrer Story oder Ihrem Chat".

Functionally, the image generator stands out from earlier solutions. Combined with the large language model Muse Spark, the system can, according to The Verge, "think prompts through logically, search the web, and plan ahead for image generation." Wang put it this way: the AI will "think through your prompt, research the web for it, and plan before generating anything." This means Muse Image can also independently create diagrams, functional QR codes, animated GIFs, entire web pages, and simple visual games.

Features and Technical Innovations

At the same time, the system has a self-optimization capability. According to Meta, this property did not develop deliberately but emerged "during training." When asked, the system independently suggested improvements during a test, asking, for example: "Would you like me to adjust the background further, e.g., sunset or more waves?" A typical editing recommendation was: "Switch the flower to a tulip."

In addition to free-form generation, Muse Image offers ready-made presets. These include adjusting the lighting of a photo, removing objects or people from the background, and tailoring advertisements. A special feature allows rooms to be redesigned based on images; this is primarily intended to be linked to Facebook Marketplace. Meta demonstrated the capability in a promotional video in which users can visualize a used sofa in their garage – the company puts it this way: "so lässt sich beispielsweise vorab prüfen, wie eine gebrauchte Couch im eigenen Wohnzimmer wirken würde". The feature was internally known by the codename Mango.

Perhaps the most sensitive new feature concerns how content from other users is handled. Muse Image allows users to AI-manipulate the images of other Instagram users, provided their profiles are publicly visible. For this, "a simple mention of a user via @ in the prompt" is enough, as the company explains. The AI can thus alter a person's appearance or place their images into a different context. It is also possible to tag public profiles in AI prompts and generate images from them using other people's photos, Reels, or audio.

Reactions to this feature have been correspondingly critical. On the X platform, the approach has already been described as a "Privatsphäre-Minenfeld mit Detonationsgarantie". Critics complain that "reale Personen ohne explizite vorherige Zustimmung in generierte Fotos hineingezogen werden". Another concern involves the audio function: when activated, anyone can use the full original audio recordings or excerpts of them to create content in apps and on websites with Meta AI.

Data Protection and the Debate Around Public Profiles

Meta, for its part, points out that account holders can use settings to disable this kind of co-optation of their images. Those who want to protect their privacy, however, must take action themselves, since the feature is enabled by default for public accounts. As the company explains, it thereby maximizes "die verfügbare Bild-, Video- und Audiobasis" and draws on billions of public photos, videos, and identity data stored on its platforms.

Anyone who wants to prevent reuse must go to their own profile in the Instagram app, tap the menu with the three lines in the top right, and open "Settings and activity." Then scroll to the "Share and reuse" section and disable the individual toggles – separately for posts, Reels, and original audio recordings. However, this only stops future generations; already-created AI images remain and are not deleted. The same applies if an account is subsequently switched to private, as the company concedes.

Critics point out that the invisible watermark "Content Seal" introduced by Meta does confirm the AI origin of an image but does not prevent an unwanted image from being created in the first place. The company counters that "Filter gegen klar schädliche Inhalte greifen". In Europe, according to the report, the feature is likely to be watched closely with regard to the GDPR and the protection of biometric data.

U.S. Market Launch and Outlook for Europe

The market entry is taking place in the United States and via WhatsApp; there, Muse Spark itself said: "Die Bild-Funktion wird bald freigeschaltet". In addition, according to Meta, 30 new AI effects for WhatsApp and Instagram are being rolled out, initially available only in the United States. For Europe and Germany in particular, the company did not name a specific timeline – the feature has not yet been integrated.

At the same time, the new system replaces an older generation. As Meta explains, image and video functions previously ran on external models such as Midjourney and Black Forest Labs; Muse replaces the previous Llama model line in this segment. The advertising industry is also being brought in: Meta's Advantage+ tools allow brands to work with personalized ad creatives. This creates a direct bridge between the image model and the company's core business.

Business Model: Cross-Subsidization Through Ad Revenue

The economic logic behind this is understandable: Meta has vast stocks of user-generated data and can cross-subsidize the costly AI business with advertising revenues from its own social networks. Despite what is perceived as an unclear AI strategy, the company continues to invest in its AI infrastructure. In recent months, Zuckerberg has also poached top talent from competing AI labs to strengthen his own team.

The privacy debate is being further fueled by a specific incident from the report's own testing. According to it, Meta AI generated a LinkedIn-suitable photo of the author together with Donald Trump – without any watermark indicating it was a fake. Another request, to depict the author's photo together with Trump in swimwear, was rejected by the system, which instead suggested summer business-casual attire.

In another self-test, Muse Spark delivered sticker suggestions modeled on the design of "My Melody," as well as alternative image ideas for a vacation composition after an initial request was rejected due to a copyright-protected image from a travel blog. These and similar experiences underscore that the model opens up creative possibilities on the one hand, but on the other raises unfamiliar questions of control and copyright.

An outlook on further functions is already foreseeable. As Meta announced, an AI video generator called Muse Video is to complement the image lineup; in addition to already released applications such as the AI assistant Creator or the Pocket app, the video model is already in development. Until this offering is available, Meta wants to roll out Muse Image to additional countries, without naming a date for Europe so far.

What Users Can Do

In parallel, Meta is working on the gradual introduction of the new settings and formulations. Anyone checking now may therefore not yet find all options or exactly the new labels, according to the company. The full range of functions, including the audio option described in settings as "Allow others to creatively work with and reuse your original audio recordings on Meta AI," will be activated piece by piece.

In summary, Meta is positioning Muse Image as a creative tool for end users, advertisers, and Marketplace sellers alike. At the same time, the company is shifting responsibility for protecting private content more strongly onto users, since the default setting for public accounts allows reuse, and the resulting AI images are not automatically removed even after an opt-out or a switch to a private account.

Data protection authorities in Europe are watching the development closely. Since the system can process biometric features, the GDPR and national regulatory frameworks could come into play here, especially once the product is rolled out in Germany and the EU.

Work on AGI remains the overarching goal, according to the company. Mark Zuckerberg had previously formulated a "personal superintelligence" as his vision. In this reading, Muse Image is meant to be an initial visible building block of a long-term AI strategy in which creative tools are embedded directly into existing social platforms and advertising ecosystems.