Microsoft Launches MAI-Image-2 and Challenges Google and OpenAI

Microsoft Unveils MAI-Image-2, Its New AI Image Generator

Five months after the launch of MAI-Image-1, Microsoft announced on March 19, 2026, the release of MAI-Image-2, an enhanced version of its artificial intelligence image generator. Presented by the Microsoft AI Superintelligence (MSI) team, this new iteration promises significant advancements in photorealism, reliability of text embedded in images, and richness of generated scenes. Users can already test the model on MAI Playground, and its deployment is underway on Copilot and Bing Image Creator.

Collaboration with Creatives for Tangible Improvements

To develop MAI-Image-2, Microsoft worked closely with photographers, designers, and content creators. The model focuses on three main areas:

• Enhanced Photorealism: Natural light rendering, skin textures, and environments are improved to reduce post-production retouching time.
• Text in Images: Text generation is more reliable, facilitating the creation of infographics, typographic posters, and diagrams.
• Detailed and Complex Scenes: Tailored for cinematic or surreal compositions, these scenes offer increased visual richness.

API access is currently limited to select clients but will soon be available for developers through Microsoft Foundry. Users interested in commercial use can already submit a request for access. Deployment in France is ongoing and may take a few days or weeks.

Microsoft Makes Progress in a Competitive Market

On the Arena ranking, formerly known as LMArena, which evaluates image generation models through blind voting, MAI-Image-2 ranks fifth. This result allows Microsoft to claim the position of the third-largest lab in this field, behind Google and OpenAI. This progress is notable compared to MAI-Image-1, which ranked ninth last October. However, Microsoft will need to continue innovating to close the gap with leading models such as GPT-Image and Nano Banana.

This launch coincides with an active week for AI image generation, where Midjourney also unveiled version 8 of its model in alpha. With MAI-Image-2, Microsoft asserts its commitment to developing its own generative AI capabilities, gradually moving away from the OpenAI models it previously used.