AI image and video generators are the kind of AI products that go viral frequently, effectively becoming marketing tools for their creators. OpenAI’s 4o Image Generation model and the Sora 2 video model are such examples. ChatGPT went viral earlier this year for its image generation abilities, helping OpenAI sign millions of new ChatGPT users. The recently launched Sora app became the most downloaded iPhone app in less than a week. Google’s Nano Banana image generator is another good example, having helped Gemini become the most popular iPhone app in the weeks preceding the Sora release. Seeing Microsoft AI announce a new AI image generation model developed in-house (MAI-Image-1), and tease its exciting performance and speed, is no surprise.
MAI-Image-1 might not have a name as cool as Google’s Nano Banana, but Microsoft AI is positioning it as a state-of-the-art model that can compete against the best alternatives out there. Microsoft’s AI lab noted in a blog post that MAI-Image-1 currently ranks in the top 10 text-to-image models on LMArena, which is impressive. LMArena is a testing ground for unreleased AI products where users can compare different models before their public release.
When will MAI-Image-1 be available in Microsoft products?
Microsoft says it developed MAI-Image-1 with the goal of “delivering genuine value for creators.” The company put “a lot of care into avoiding repetitive or generically stylized outputs,” prioritizing “rigorous data selection and nuanced evaluation focused on tasks that closely mirror real-world creative use cases” and taking into account feedback from creators. That’s a lot of marketing speak in Microsoft’s announcement that does little to explain what MAI-Image-1 is good for. The company says that MAI-Image-1 “excels at generating photorealistic imagery,” including lighting effects, landscapes, and “much more.” Microsoft also teases that MAI-Image-1 is faster than some of its direct competitors, without offering any benchmarks to back those claims.
Still, MAI-Image-1 coming in the ninth place in LMArena’s top 10 for text-to-image models, as of this writing, indicates that Microsoft’s in-house AI image generator is appreciated by testers. LMArena’s top 10 includes several models from Google, ByteDance, and Tencent. Nano Banana is ranked second. The images above were probably generated with MAI-Image-1, as they were included in the official announcement. They don’t feature any watermarks that would indicate their origin.
Microsoft says it will take into account feedback from early users before releasing MAI-Image-1 in Copilot and Bing Image Creator. It’s likely that MAI-Image-1 will be available in other Microsoft products that feature built-in AI capabilities, as Microsoft AI teases MAI-Image-1 to be the “step on our journey and paves the way for more immersive, creative, and dynamic experiences inside our products.”
