Microsoft AI has unveiled its first text-to-image generator developed entirely in-house, MAI-Image-1. Microsoft AI touts its new model’s performance, especially concerning generating photorealistic images.
MAI-Image-1 debuted in the top 10 of LMArena’s text-to-image AI generator leaderboard, scoring a solid 1,096 preliminary mark. This puts it behind models like Google’s Gemini 2.5 (“nano banana”) Imagen 4.0 Ultra, but ahead of AI image generators like Dall-E and Stable Diffusion.
The fact that MAI-Image-1 is outpacing Dall-E is notable, given that Dall-E’s creator, OpenAI, has historically had very close ties with Microsoft, including a multi-billion dollar partnership. However, more recently, OpenAI and Microsoft’s relationship has appeared a bit rocky, even though the two companies are still strategically partnered on AI innovation and safety.
As The Verge reports, Microsoft recently began using Anthropic AI in certain Microsoft products. And of course, Microsoft is clearly investing heavily in developing its own AI these days, as MAI-Image-1 demonstrates.
Microsoft AI describes MAI-Image-1 as the “next step” in its AI journey, and claims that the text-to-image generator will contribute to developing “more immersive, creative, and dynamic experiences” inside Microsoft’s products.
Microsoft solicited the help of professionals in various creative industries when training MAI-Image-1, which the company says helped make the model so powerful at launch and will make it more useful for real-world professionals.
“For example, we prioritized rigorous data selection and nuanced evaluation focused on tasks that closely mirror real-world creative use cases — taking into account feedback from professionals in the creative industries,” Microsoft says. “This model is designed to deliver real flexibility, visual diversity and practical value.”
The new AI image model is particularly good at generating photorealistic images, per Microsoft, including realistic lighting (including bounce light and reflections), landscapes, and “much more.” The tech giant touts its new model’s speed, which Microsoft claims surpasses many larger models on the scene.
To ensure that MAI-Image-1 is safe when it is officially released in Microsoft Copilot and Bing Image Creator “very soon,” Microsoft is initially launching its new text-to-image model in LMArena, where users can test it now.
“We are committed to ensuring safe and responsible outcomes,” Microsoft promises.
The early images generated by Microsoft’s new AI image model, which are featured throughout this article, look quite impressive. For photographers, it’s not obvious that this is a good thing.
Image credits: Microsoft