MAI-Image-2.5

MAI-Image-2.5 is Microsoft's text-to-image model from the MAI Superintelligence Team, launched on May 26, 2026. It debuted at No. 3 on the Arena text-to-image leaderboard and is positioned as a step change over MAI-Image-2 with major improvements in text rendering, stylized illustration, and commercial imagery.

MAI-Image-2.5 was built by the MAI Superintelligence Team inside Microsoft. It continues the in-house MAI-Image series that started with MAI-Image-1 and MAI-Image-2, Microsoft's first-party text-to-image models.

MAI-Image-2.5 generates images from text prompts across a wide range of styles. It is particularly strong on in-image text rendering for posters, labels, and packaging, on stylized illustration, and on commercial product photography with believable lighting, scale, and spatial structure.

Microsoft describes 2.5 as a step change in quality over MAI-Image-2, with the largest gains in text rendering, stylized illustration, and commercial imagery. On the public Arena leaderboard, 2.5 sits well ahead of MAI-Image-2 at launch.

MAI-Image-2.5 launched at No. 3 on the Arena text-to-image leaderboard on May 26, 2026, behind models from Google and OpenAI and ahead of its predecessor MAI-Image-2.

At launch, MAI-Image-2.5 is available in preview on Arena, and Microsoft has said it will roll out to the MAI Playground and Microsoft Foundry within two weeks of the May 26, 2026 announcement.

Sharper text rendering is one of the headline upgrades. Words on posters, labels, and packaging keep their spelling and shape, and layouts stay structurally intact as copy gets longer or more decorative.

Yes. Microsoft highlights commercial imagery, product shots, and packaging as the areas where 2.5 improved most, with stronger material truth, lighting, and brand polish suited to landing pages, ads, and client decks.

The MAI-Image-2 model card describes the family as a diffusion-based text-to-image architecture trained with a flow-matching loss. MAI-Image-2.5 continues that lineage as the latest release in the MAI-Image series.

ChatGPT Images 2.0 leads on reasoning-driven generation, multi-image batches with character continuity, and output up to 2K. Nano Banana focuses on fast iteration with strong real-world grounding. MAI-Image-2.5 sits between them, with its largest reported strengths in in-image text rendering, stylized illustration, and commercial product and packaging imagery. Pick MAI-Image-2.5 when text legibility and brand polish are the most important parts of the brief.

Open Morphic Copilot, describe the image, and pick MAI-Image-2.5 from the image model picker. It sits alongside ChatGPT Images 2.0, Nano Banana, Flux, and the rest of the catalog, so you can switch in one click when text rendering or commercial polish is central to the brief.