Meta launches muse image: The new "reasoning" AI generator built into WhatsApp and Instagram

Meta Platforms Inc. has officially rolled out Muse Image, its first entirely in-house AI image generation model. Developed by the newly formed Meta Superintelligence Labs, Muse Image departs from traditional generators by utilizing an “agentic tool-use” framework. Rather than just mapping phrases to pixels, the model acts as an agent—actively writing code, executing scripts, and browsing the open web in real-time to drastically improve factual accuracy.

The model is currently live and free to use inside the Meta AI standalone app, meta.ai on the web, WhatsApp chats, and via more than 30 new interactive Instagram Stories effects in the United States and select international markets. Plans are underway to expand the algorithm to Facebook and Messenger in the coming weeks.

The Core Technical Shift: “Deliberate Reasoning” over Best-of-N

Unlike standard AI generators that generate multiple media files simultaneously and filter for the best result (a method known as Best-of-N), Muse Image utilizes a native “deliberate reasoning” approach during test-time compute. This architecture allows the model to scale its performance depending on the hardware power allocated to it.

Advanced Features and Capabilities

Flawless Built-in Text: Muse Image remedies a multi-year industry problem by generating readable, correctly spelled typography on banners, invitations, and graphics.

Scan-Ready QR Codes: During reinforcement learning, the algorithm learned to independently execute Python code to overlay fully functional, scannable QR codes directly into generated flyers or promotional layouts.

Granular Markup Editing: Users can upload a photo or sketch, draw a circle over a specific region, and issue conversational instructions (e.g., “remove the fog,” “swap the background weather,” or “change the flower to a rainbow gradient”) without altering the rest of the image.

While the creative capabilities of the model have drawn praise, Meta’s rollout has immediately triggered intense user backlash across social networks over data privacy boundaries.

The primary controversy stems from a feature that allows users to type in a public Instagram handle as part of a prompt to modify or remix their public images. Furthermore, Meta’s corporate policy stipulates that everyday users will not receive automated notifications when their public photos are pulled into third-party AI generations.

The Opt-Out Reality: Public Instagram accounts are opted into the Muse Image generation ecosystem by default. To protect personal photography, users must manually navigate to their privacy settings to toggle the feature off.

Corporate Deployment & What’s Next: Muse Video

Meta is actively positioning the Muse family as a core pillar of its social commercial strategy. Within the next month, the image model will integrate directly into the Advantage+ creative marketing suite, allowing small businesses to build product mockups or alter marketing materials automatically.

Additionally, Meta AI’s “Shopping Mode” will allow consumers to upload a photo of an empty room and overlay realistic, accurately scaled furniture pulled from real business catalogs.

Meta’s new Muse family of models is rolling out across 2026 with a big push into multimodal AI.

Muse Spark 1.1 launched between April and July 2026 and is already active. It powers the “Thinking Mode” inside Meta AI as a multimodal language reasoning model.

Muse Image went fully live on July 7, 2026. It’s now deployed across WhatsApp, Meta AI, and Instagram Stories in the US for image generation.

Coming next is Muse Video, slated for a late 2026 release. A preview is already live. It’s a native audio-supported AI clip generator and currently ranks 3rd on the public Arena AI leaderboards.