New Delhi: Meta this week began rolling out Muse Image, the first image-generation model built by Meta Superintelligence Labs, making it available inside the Meta AI app, and parts of Instagram and WhatsApp. The company said the model also powers more than 30 AI effects for Instagram Stories and image-generation in direct chats with Meta AI in WhatsApp, initially in a limited set of countries.
The rollout has drawn attention for one feature in particular: the ability to type an “@” followed by a public Instagram username to pull photographs from that account into an AI-generated image.
Meta says users can switch the setting off. Lawyers working on technology policy say the burden of that switch, and what it does and does not undo, is the point of contention.
The Ministry of Electronics and Information Technology said on Thursday that it would examine the feature.
Also Read: Meta’s cloud play will spice up Ambani-Adani rivalry
What the model does
According to Meta’s announcement, Muse Image is designed to take instructions in conversational language and produce images that can be downloaded or posted to a feed, story or chat. The company lists erasing an object from a photograph, placing a user in front of a landmark, and generating a functional QR code among the tasks the model can perform.
Meta says text inside generated images renders legibly, allowing for how-to guides and infographics.
The company describes a presets panel of suggested prompts that can restore an old family photograph, apply a hairstyle, or render a user as a claymation character or a 16-bit video game character. A separate feature lets a user photograph a room and ask Meta AI to redesign it using products drawn from the web or Facebook Marketplace.
Meta says the model works alongside Muse Spark, its assistant model, and takes several steps before producing an output. In the company’s description, the model plans a layout, retrieves real-time web context, and blends multiple visual references before rendering. Users can circle, sketch or annotate directly on a generated image to request changes, and the model retains the context of the conversation across edits.
Meta says the tool is free for everyday use and is included in the company’s subscription plans for higher volumes. It says Muse Image will expand to Facebook and Messenger, to more surfaces on Instagram and WhatsApp, and to advertisers and agencies through Advantage+ creative in the coming weeks.
A video model, Muse Video, is in development.
The ‘@’-mention feature
In its announcement, Meta writes that a user can “@-mention Instagram accounts in the Meta AI app to bring specific Instagram profiles right into your images,” and that “tagging a username lets Meta AI use public photos to build a visual that’s ready to post.” The company adds that users “have control over how your content can be tagged for AI creation with an easy setting to turn this feature off at any time.”
The design places the control after the fact. Public accounts are available as visual references unless the account-holder locates the setting and disables it.
Mishi Choudhary, founder of the Software Freedom Law Centre, said the design shifts responsibility from the platform to the individual.
“If public Instagram accounts can be used as visual references by default, that is a product design choice that shifts the burden onto users to restrict or opt out of reuse,” Choudhary said. “It also increases the risks of impersonation, targeted harassment and fraud, particularly where realistic synthetic images can be generated at scale.”
Choudhary said the consent question turns on the difference between publishing a photograph and licensing it for downstream use.
“Publicly sharing a photograph for one purpose should not automatically make it available for every downstream AI use,” she said. “Platforms introducing these capabilities should be expected to give users clear notice, meaningful control and effective safeguards by design.”
Choudhary pointed to Meta’s own Help Centre documentation, which she said sets out two limits on the opt-out.
“According to Meta’s own Help Center, users will not be notified when AI generated content is created using their Instagram content, and changing the setting prevents future use but does not remove content already generated,” she said. “That raises important questions about user awareness, control, and the adequacy of an opt-out design.”
Under that design, an account-holder who has not opened the setting has no signal that an image drawing on their photographs exists. An account holder who opens the setting and disables the feature stops subsequent use, but images generated before that point remain.
Meta’s announcement does not state whether users are notified when their content is used, and does not address the status of images generated before a user changes the setting.
The government’s position
Asked about the feature at an event on Thursday, Ministry of Electronics and Information Technology Secretary S. Krishnan told reporters that the government would examine it. “We will have to look at it with reference to the legal framework and also representation,”
He did not specify which provisions the ministry would apply, or whether the government had received a representation on the feature.
Jaspreet Bindra, co-founder of AI & Beyond, said the release signals a change in how image models operate rather than an improvement in output quality alone.
“The launch of Muse Image marks a fundamental pivot from static text-to-image synthesis to what I call ‘agentic media creation’,” Bindra said. “Meta isn’t just mapping words to pixels anymore; they have introduced an architectural shift where the model behaves as an autonomous agent invoking search, writing code to ensure precision like scannable QR codes, and dynamically self-refining its output.”
Bindra said the effect on the industry is a move away from single-prompt generation. Content creation, he said, “will move away from brute-force prompting into an era of conversational, contextual co-creation deeply embedded within the social platforms where billions of people already live and interact.”
On the commercial use of such models, Bindra said the change is in the cost of producing routine visual assets. Producing product mock-ups, campaign graphics and localised social media material “required significant capital and long turnaround cycles”, he said.
Models that handle spatial design, image editing and text layout allow businesses to “transition to a zero-marginal-cost model for everyday creative visual assets,” moving them “from investing in heavy standalone execution software to strategically automating their baseline content supply chains.”
For creators and small companies, Bindra said, the features that matter are multi-photo blending and personalised effects, which allow teams to “test narrative angles, design precise infographic backdrops, and run contextual visual experiments across platforms like Instagram and WhatsApp without structural overheads.”
Asked whether designers face displacement, Bindra said: “AI is not going to replace graphic designers, but graphic designers who use AI will replace those who do not.” He said such systems handle “erasing background photobombers, complex object tracking, or generating initial baseline iterations,” leaving designers to work on “conceptual depth, strategic narrative, and emotional resonance.”
Bindra said the privacy question is the one to watch. Models that draw on public data or user profiles to generate likenesses create “an unprecedented privacy friction point, shifting the burden entirely onto consumers via an opt-out framework”, he said.
He called for “invisible watermarking systems or ‘Content Seals'” and said users and companies “must maintain strict governance, verifying data provenance and ensuring meaningful consent mechanisms”.
What users can check
Meta’s Help Centre, accessible from within the Instagram app, sets out the setting that governs whether an account’s public photographs can be used when another user tags that account in Meta AI.
The setting applies going forward. It does not apply retroactively, and it is not accompanied by a notification when an account’s content is used. Meta has not published figures on how many accounts have changed the setting, or on how many images generated to date have drawn on tagged accounts.
(Edited by Ajeet Tiwari)
Also Read: Meta, NYU study finds video, not text, is better at teaching AI how the physical world works