Meta Muse Image AI Model: Launch Sparks Privacy Row 

The Meta Muse Image AI model landed this week, and within hours, it had done two things at once: shown off some genuinely impressive image-editing chops, and reopened old wounds about how the company treats its users’ photos.

Meta rolled out Muse Image on Tuesday, July 7, 2026, marking the first image-generation model from Meta Superintelligence Labs, the AI division led by Alexandr Wang. For a company that spent years leaning on outside vendors, this is a deliberate pivot. Meta had previously relied on third-party models from Midjourney and Black Forest Labs to power image generation inside its apps. Muse Image is the in-house replacement.

The model was reportedly built under the internal codename “Mango,” and it arrives roughly three months after Meta Superintelligence Labs’ first major release, the Muse Spark language model, which succeeded the Llama family in April. That timeline matters. It tells you Meta isn’t dabbling, it’s stacking releases quickly to catch up with rivals like OpenAI and Google.

How It Actually Works

Rather than turning a prompt straight into a picture, Muse Image behaves more like an agent working through a problem and invokes search and coding tools to sharpen accuracy, refines its own output, and improves by throwing more compute at the task during generation itself. In practical terms, that means it can plan a layout, incorporate real-world references, and stitch together multiple photos before settling on a final image.

The Instagram Tagging Problem

Here’s where things get uncomfortable. Meta has built in an @-mention feature that pulls public Instagram photos directly into AI-generated images and the setting is enabled by default, meaning users have to actively opt out if they don’t want their public photos used this way. Given Meta’s history, the company paid a then-record five billion dollar FTC fine in 2019 after the Cambridge Analytica scandal exposed how user data had been harvested without consent, it’s not surprising this feature is already drawing criticism.

Where It Stands Against Rivals

On the performance side, Meta isn’t claiming the crown outright. Internal benchmarks show Muse Image trailing OpenAI’s GPT Image 2 model, but outperforming Google’s Nano Banana 2 on single and multi-image editing tasks. That’s a candid admission for a company that usually leads with confidence, and it suggests Meta is positioning this as a strong second-place contender rather than an outright leader.

Where You’ll Find It and What’s Coming Next

Muse Image is live today in the Meta AI app, on meta.ai, and on Instagram Stories in the US, with WhatsApp access rolling out in select countries; Facebook access is coming soon. On the business side, advertisers and agencies are expected to get access to Muse Image-powered ad variants within the Advantage+ tool in the coming weeks, which lines up with Meta’s broader push to squeeze more ad revenue out of its AI investments.

To address concerns around authenticity, Meta has added Content Seal, an invisible watermarking system that keeps a hidden provenance marker embedded in images even after cropping, compression, or screenshotting.

A video counterpart, Muse Video, is already in the works, built on the same underlying architecture. For now, though, all eyes are on how Meta handles the Instagram photo controversy, because that story is likely to shape how the Meta Muse Image AI model is remembered, far more than any benchmark score.