Meta opens agentic Muse Image model to developers at $0.01 per generation

Meta's Muse Image is now available to developers directly on the Meta Model API and is priced at $0.01 per image for production volumes. [2] The model functions as a reasoning model that maps out a full layout and plans before generating or editing images. [2] Instead of rendering a request straight away, Muse Image thinks about it first, mapping out a plan and pulling real references from the web. [1]

The model allows developers to decompose, annotate, and revise specific parts of an image in a single call. [2] During training, the model's ability to self-refine emerged on its own, allowing it to reflect on drafts and improve them through local edits or tool use. [2]

Muse Image’s aggressive $0.01 pricing allows workloads that were previously restricted by frontier pricing to become viable at scale, specifically targeting applications like ad variant generation and per-user personalization. By bringing reasoning and self-refinement into the image generation process, Meta is pushing agentic capabilities beyond text, allowing the model to generate and run code where precision is required. While the model has enabled creative experiences within Meta AI since July, this API release shifts architectural control to enterprise developers. What the sources don't address: How the model's latency is impacted when executing multi-step reasoning and self-refinement loops before returning a final image.