Microsoft has introduced MAI-Image-2, a new in-house AI model for generating images from text descriptions. It will soon replace the first version of the LLM, which it has already surpassed on the image generator testing platform.

MAI-Image-2 outperforms its predecessor in several metrics. It more accurately renders natural lighting, skin tone, and other details, and is better suited for generating images with text (infographics, diagrams, slides, etc.).

Microsoft also announced improvements in photorealism.

The most exciting creative works are unusual, cinematic, hyper-detailed pieces. MAI-Image-2 is designed for such a space: surreal concepts, intricate compositions, and ambitious worlds that turn imagination into images.

In creating MAI-Image-2, the company drew on the experience of photographers and designers. The AI model is positioned as a tool for creative people who want to spend more time bringing their ideas to life rather than processing images.
Microsoft

MAI-Image-2 took third place in the Arena.ai ranking, behind only Google gemini-3.1-flash-image-preview and OpenAI gpt-image-1.5-high-fidelity. For comparison, MAI-Image-2 is currently in ninth place on the same chart.
🫟 The new image generator is now available via MAI Playground or through the API on Microsoft Foundry.

It will later be included in Copilot and Bing Image Creator.

The service does not work with Russian IP addresses)

Get it here