Meta’s new Muse Image model isn’t your typical text-to-pixel generator. Developed by the company’s super intelligence lab, it behaves more like a design agent, planning out compositions before it draws.
It works in tandem with the recently released Muse Spark language model. Before generating an image, Muse Image first sketches out a composition plan. If the prompt needs factual information, it searches the web for accuracy. If precise typography is required, it can render clear, readable text directly in the image — and it can even build functional QR codes from custom prompts.
Users can also circle and annotate directly on the canvas to tell the model what context to pay attention to and fine-tune local details, without having to regenerate the entire picture.
Muse Image is now live in the Meta AI app. Everyday creations are free to use, while more advanced features require a paid subscription. It will also roll out to commercial systems like Advantage+ ad creative over time.