Meta Launches Muse Image AI, Its First In-House Image Generation Model

Meta has released Muse Image, its first proprietary AI image-generation model, marking a significant step toward building an in-house AI creative stack and reducing dependence on third-party tools like Midjourney and Black Forest Labs. Originally codenamed Mango, the model was developed under Meta Superintelligence Labs, the division led by Alexandr Wang — the Scale AI founder who earlier oversaw the April launch of the Muse Spark large language model, which replaced Meta's longstanding Llama family.

Muse Image is rolling out across Meta's major surfaces: free access via the Meta AI app and website, WhatsApp direct messages, and Instagram Stories. Heavy users and creators who exceed free-tier limits must subscribe to one of Meta's monthly plans, introduced in May, or purchase a Meta One subscription. The tiered model signals Meta's push to generate subscription revenue alongside its dominant ad business.

On the advertising side, Muse Image is being integrated into Meta's Advantage Plus platform, enabling brands to automatically generate on-brand image variants for ad campaigns with fewer manual iterations. Meta says agencies and advertisers can expect Muse Image-powered variants within weeks — a move that could meaningfully accelerate ad creative production at scale.

In internal benchmarks, Muse Image trails OpenAI's GPT Image 2 but outperforms Google's Nano Banana 2 model on image editing tasks, both single and multi-image. Google's Nano Banana became a consumer hit after its release last fall, giving both Google and OpenAI a head start in the image-gen race. Meta also teased a forthcoming Muse Video model, describing it as competitive in prompt adherence, visual fidelity, and temporal consistency, though no release date was given. Muse Image will expand to Facebook, Messenger, and additional Instagram and WhatsApp surfaces later in 2026.