Muse Image 1.0: Meta's agentic AI image model

Muse Image 1.0 (also called Muse Image) is the first image model from Meta Superintelligence Labs. What sets it apart is that it's agentic: rather than turning a prompt straight into an image, it reasons about what you asked for, gathers real-world context, and plans the result before generating. That extra thinking shows - Muse Image holds the No. 2 spot on the LMArena leaderboard for text-to-image, single-image editing, and multi-image editing by human preference.

Muse Image 1.0 pairs with Meta's Muse Spark language model to work like an agent. It interprets your prompt, pulls real-time context from the web, searches for reference imagery, plans the layout and composition, and even self-refines - invoking search and reasoning tools to get details right before a single pixel is drawn. You can dial reasoning strength between low and high, so quick ideas stay fast while complex, real-world-accurate scenes get the full treatment.

Muse Image 1.0 is built to compose, not just generate. Bring several reference photos - a subject, a setting, a style - and it reasons about how they fit together, then blends them into a single high-quality image. Generate 1, 2, 4, 6, 8, or 10 options per request across aspect ratios from square 1:1 to wide 16:9 and tall 9:16, and export as PNG, JPEG, or WebP. It's image generation that understands your inputs instead of overwriting them.

Muse Image 1.0 is available in Picsart's AI Playground, where you can generate with it directly and compare its output against 170+ other AI models from a single prompt - no setup or model configuration required. And you can reach it whichever way you work: on the web, in the Picsart desktop app, or built straight into your own projects via CLI, MCP, REST API, and SDK - with no third-party API keys or separate subscriptions to manage.