Meta has unveiled Muse Image, its most advanced AI-powered image generation model to date, while also previewing Muse Video, marking the first media generation models developed by Meta Superintelligence Labs. The launch expands Meta's generative AI portfolio with agentic capabilities designed to improve image and video creation through reasoning, tool use and iterative refinement.

Muse Image is now available through the Meta AI app, meta.ai, Instagram Stories in the US, and WhatsApp in select countries, with support for Facebook expected soon. Muse Video will initially be rolled out to creators and Meta AI users in the coming months.

Unlike conventional text-to-image models that directly convert prompts into images, Muse Image functions as an AI agent capable of planning, refining and enhancing its outputs. The model can invoke external tools such as search and coding, enabling it to verify information, improve image accuracy and refine generations through additional computation at inference time.

Meta said the model faithfully follows complex instructions, supports precise image editing, composes images from multiple reference inputs and leverages contextual understanding derived from Instagram to generate more socially relevant visual content.

A key differentiator of Muse Image is its integration with Muse Spark, allowing both models to collaborate by sharing tools and planning jointly to support more sophisticated media generation workflows.

The company has also equipped Muse Image with coding capabilities that enable it to generate and execute code for creating accurate charts, plots and QR codes. During reinforcement learning, the model learns to use rendered figures to improve image quality and accuracy.

By combining coding with media generation, Muse Image and Muse Spark can also create animated GIFs, interactive websites featuring embedded AI-generated images, and simple visual games, expanding the potential use cases beyond traditional image generation.

Meanwhile, Muse Video, built on the same pre-training foundation as Muse Image, is designed to deliver high visual fidelity while supporting native audio generation, positioning it as a next-generation multimodal video creation model.

The announcement reflects Meta's growing focus on agentic AI, where models go beyond content generation to reason, use external tools and iteratively improve outputs. By integrating image, video and coding capabilities into a unified workflow, the company aims to provide creators, developers and businesses with more capable AI systems for digital content creation.

The launch of Muse Image and the upcoming release of Muse Video underscore Meta's continued investment in generative AI as competition intensifies among technology companies developing advanced multimodal foundation models for creative and enterprise applications.