Muse Image is a proprietary image generation model developed by Meta Superintelligence Labs (MSL) and released on July 7, 2026. Marking a shift from Meta's traditional open-weights approach, Muse Image is closed-source and operates natively inside Meta's core applications, including Meta AI, WhatsApp, and Instagram Stories. It functions as a core visual engine for the company's automated advertising and consumer creation features.
Unlike standard text-to-image models that map words directly to pixels, Muse Image is designed as an "agentic" creator. In tandem with the Muse Spark reasoning model, it plans layouts, calls external tools, executes code, and performs web searches to gather real-time context before producing an output. This approach allows the model to handle complex instructions and multi-object spatial reasoning, such as writing and compiling functional code to embed actual, scannable QR codes directly into generated artwork. It is also capable of blending multiple reference photos into a single composition while maintaining coherent lighting and perspective.
For consumer workflows, Muse Image supports direct-manipulation editing, where users can sketch or circle a region on an image and describe modifications in natural language. The model also features deep social context integration, allowing users to @-mention public Instagram accounts to use public photos as visual references in prompt generation. This setting has sparked user privacy debates, as public profiles are opted into this remixing feature by default. To balance safety and provenance, Meta embeds a hidden digital watermark, or "Content Seal," inside the generated images to trace their origins.