Muse Image

• 🆕 Meta's image generation and editing model, served via Meta Model API.
• 🧠 A reasoning model: plans full layout before drawing anything.
• 🎯 Anchored composition keeps edits consistent across repeated generations.
• 🖼 Accepts text prompts plus reference images for editing and composition.
• 🔧 OpenAI-style image generate and edit endpoints; base64 image output.
• 📏 Multiple output sizes, including wide formats such as 1536 by 1024.
• 💬 Self-reflects mid-chain-of-thought, patching drafts or regenerating them.
• 🌐 Can switch to tool use when factual accuracy matters.

Meta

Meta Platforms is an American multinational technology company headquartered in Menlo Park, California. Best known for operating Facebook, Instagram, WhatsApp, and Threads, Meta has become one of the most influential players in open-weight AI research through…

Muse Image is Meta's image generation and editing model, exposed through the Meta Model API under the identifier for its 1.0 release. Unlike conventional diffusion-style endpoints that map a prompt straight to pixels, Meta describes it as a reasoning model: before drawing, it inspects any input frames and maps out the full layout — what goes where, how many elements there are, and how they relate to one another. That planning step is the model's defining architectural idea rather than a post-processing add-on.

The API surface covers three tasks: generating from text, editing existing images, and composing across multiple reference frames. Requests follow OpenAI-compatible image generation and image edit calls, with results returned as base64 data that clients decode to bytes, and with size arguments for formats such as 1536 by 1024 used in multi-panel comic workflows. Meta's own documentation walks through passing several reference files into a single edit call to keep characters and scenery stable across panels, which is what its anchored composition behavior is meant to support.

Meta also reports an emergent behavior it says was not explicitly designed: inside its chain of thought, Muse Image critiques its own drafts and revises them — making a local fix when a small detail is off, regenerating entirely when larger parts are wrong, or invoking tools when factual accuracy is at stake.

This catalog entry lists no earlier Muse Image generation, so there is no same-family predecessor to compare against; the model sits alongside Meta's broader Muse line of multimodal reasoning systems.