Microsoft's new AI can make pictures in 14 seconds, which is faster than ChatGPT and Gemini.
Microsoft first talked about its MAI-Image-1 image generator last October. They said it would make ideas come to life faster and with better quality than any other image generator. Users could get this new software for free in November through the Bing app, search engine, and Bing Image Creator, which lets you make images with artistic lighting and a lot of detail. But the company wanted to go faster than the industry average, so it released MAI-Image-2 in mid-March as an option that could turn text into professional-looking images with reliable embedded text.
A month after the announcement, Microsoft released MAI-Image-2-Efficient, a version designed for creating images with a focus on speed and scalability. This version boasts 22% faster performance than the standard version and approximately 41% lower running costs.
MAI-Image-2-Efficient undoubtedly outperforms other AI models thanks to its response speed, which is 40% faster than the average text-to-image converter. While Google's Gemini 3 Pro Image takes 19 seconds, and OpenAI's GPT-Image-1.5-High takes just over 41 seconds, MAI-Image-2-Efficient can respond in less than 14 seconds.
Microsoft confirmed in a statement on its website that MAI-Image-2-Efficient is "the ideal production tool," designed for use when high workloads, high speed, and "precise cost control" are required. It can be used specifically to create product images, user interface mockups, brand assets, creative marketing materials, and short texts such as headlines or labels.
On the other hand, in addition to being optimized for working in real-time interactive workflows, MAI-Image-2-Efficient is also designed for three scenarios: prototyping, real-time conversations, and mass production.
Regarding its availability, MAI-Image-2-Efficient is already available on the Microsoft Foundry and MAI Playground platforms. It is also being integrated into Copilot and Bing and will be available in tools like PowerPoint later.