Meta's Muse Image lands on the API: it reasons before it renders, at a cent an image
Each call searches the web and refines the output over iterative passes, with extra evaluation of the prompt for precise elements like charts and QR codes. Generation and editing sit in one model.
Meta has opened its image generation model Muse Image to developers through the Meta Model API. Pricing starts at $0.01 per image at production volumes.
The model's claim is that it does not simply take a prompt and render. By Meta's account Muse Image behaves like an agent: each call searches the web, refines the output over iterative passes, and separately evaluates the prompt for elements that leave no room for error, such as charts and QR codes. That is precisely the known weak spot in image generation, where text, tables and codes come out illegible in most models.
The second point is that it is a single model. Text-to-image, single-image editing, multi-image editing and multi-reference composition are gathered in one place. Meta frames it this way: there is no multi-step pipeline to stitch together.
The announcement came from the Meta for Developers account and was shared by AI at Meta.
Details are on Meta for Developers.
