Meta Superintelligence Labs Ships Its First Image Model, Muse

Meta Superintelligence Labs has shipped Muse Image, the group's first media generation model built entirely in-house. Rather than positioning it purely as a text-to-image generator, Meta describes Muse as an agentic system that acts on instructions rather than simply interpreting them loosely. It follows prompts faithfully, edits existing images with precision, and can compose new outputs from multiple reference images at once. The more notable claim is that Muse can call its own tools mid-generation. According to Meta, the model can invoke search and coding tools to check facts, refine details or verify accuracy before returning a final image to the user. This puts it closer to an agentic workflow than a standard diffusion model, where the system takes intermediate steps to improve output quality rather than producing a single pass and stopping. Muse Image is rolling out now inside the Meta AI app, on Instagram Stories in the US, and on WhatsApp in a limited number of countries. Meta says a video counterpart, Muse Video, is coming soon, suggesting the same agentic approach — instruction-following, editing and tool use — will extend to video generation once it ships. The release marks a shift for Meta Superintelligence Labs, the division formed to consolidate the company's frontier AI efforts, into consumer-facing generative media. It also signals Meta's intent to build its own image and video models rather than relying solely on partnerships or open releases from other labs, positioning Muse as a direct competitor to image tools from OpenAI, Google and Midjourney within Meta's own apps.