MAI-Image-1: legacy model and current Microsoft alternatives

MAI-Image-1 is Microsoft’s legacy text-to-image model; its current documented family is MAI-Image-2.6 and 2.6-Flash. This guide preserves the original October 2025 and June 2026 MAI-Image-2.5 launch history. For new generation or editing work, verify the exact offered model, product, preview status and terms. Microsoft’s 2.6 card covers text and image inputs, but deployment controls must be checked separately; the older 2.5 limits below are version-specific.

This page preserves the established MAI-Image-1 query while separating launch history from current product guidance. Availability can vary by product, account, region, and rollout; verify the model name shown in the interface before using a result in production.

• Original role: Microsoft's first image-generation model developed in-house.
• Announcement: October 13, 2025, with a top-10 LMArena debut reported by Microsoft.
• Launch access: Microsoft said on November 4, 2025 that MAI-Image-1 was available as an option in Bing Image Creator and in Copilot Audio Expressions Story Mode.
• Current successor: Microsoft now documents MAI-Image-2.6 and MAI-Image-2.6-Flash for generation and editing; MAI-Image-2.5 is part of the earlier launch history below.
• New-work recommendation: check a currently offered MAI-Image-2.6 route when its documented generation or editing capabilities fit; keep MAI-Image-1 for legacy reproduction or research only when that exact model is available.

What MAI-Image-1 was designed to do

Microsoft described MAI-Image-1 as a fast text-to-image model focused on photorealistic imagery, lighting such as bounce light and reflections, landscapes, visual diversity, and rapid iteration. The announcement also said the training and evaluation process included feedback from creative professionals and aimed to avoid repetitive or generic styles.

Those are provider claims from the launch announcement. They do not establish that the model beats another system on a reader's product image, typography, portrait, or illustration brief. The old version of this article presented detailed strengths such as reliable kerning, stable faces, clean logos, and reduced hallucination as if Magic Hour had retained test evidence; that evidence was not present, so those claims have been removed.

What replaced it: MAI-Image-2.5

Microsoft calls MAI-Image-2.5 its strongest image model as of June 2026. That June 2026 family supports both text-to-image generation and image-to-image editing. Documented editing tasks include replacing or adding objects, updating text, removing motion blur, adapting layouts, and preserving visual consistency across iterative edits.

The MAI-Image-2.5 model card identifies text and image inputs, image outputs, and a maximum total output pixel count of 1,048,576. Either dimension may exceed 1024 as long as the total remains within that limit. That is a model limit, not a promise that every Microsoft product exposes every input, output, or control.

Current family: MAI-Image-2.6 and 2.6-Flash

Microsoft’s September 4, 2026 announcement introduced the 2.6-Flash variant and developer access through Microsoft Foundry Public Preview. It documents multi-image references, web grounding and dynamic aspect ratios for both variants, with a MAI Playground route to try them. Verify the selected product and exact model before uploading: that announcement does not establish 2.6 availability in every Bing, Copilot, PowerPoint, OneDrive or third-party account. These current-family sources were rechecked October 10, 2026.

Version-specific output limit: the MAI-Image-2.6 / 2.6-Flash model card lists 2,359,296 total output pixels, equivalent to a 1536 × 1536 square. One side may exceed 1536 provided the total stays within that count. This differs from the 2.5 limit above; it is not a guarantee that every interface offers every dimension. For a 16:9 asset, inspect the route’s allowed sizes and the downloaded file rather than assuming a square reference size is a hard per-side maximum. Licensing follows the particular product or service terms.

Where current MAI image models are available

At Microsoft Build 2026, Microsoft said MAI-Image-2.5 and its Flash variant were live in PowerPoint, rolling out on OneDrive, and landing on Microsoft Foundry. The announcement also said MAI models would be available through selected external inference providers. Check the current catalog and your product interface because staged rollouts and provider catalogs can change.

For the older model, Microsoft's MAI-Image-1 announcement records its November 2025 launch in Bing Image Creator and Copilot Labs. Do not assume Bing's current default or model menu still matches that launch state.

MAI-Image-1 vs MAI-Image-2.5

• Generation: both families generate images from text; MAI-Image-2.5 was the June 2026 successor, followed by the current documented 2.6 family.
• Editing: MAI-Image-2.5 documents image-to-image and localized editing; the MAI-Image-1 launch was centered on text-to-image generation.
• Access: MAI-Image-1 launched in selected consumer experiences; MAI-Image-2.5 launched across PowerPoint, OneDrive rollout, and Foundry, with other providers announced.
• Reproduction: use MAI-Image-1 only when the exact legacy model is available and an older output or workflow must be recreated.
• Production choice: verify the model, product controls, data terms, output limits, commercial terms, price, and regional availability for the route you actually use.

How to evaluate a Microsoft image model

• Use the same permitted reference images, prompt, aspect ratio, and output target for every candidate.
• Score prompt adherence, subject and layout preservation, typography, product geometry, faces and hands, edit locality, unwanted changes, and output dimensions.
• Keep failed attempts and count retries. A selected hero image does not show consistency or cost per accepted result.
• For editing, compare the changed region and the areas that should have stayed untouched.
• Inspect every letter, logo, number, face, hand, reflection, and repeated object at full resolution.
• Record the product, provider, model name, date, settings, latency, outside editing time, and final cost.

When to use a browser platform instead

Use Magic Hour's AI Image Generator when you want to compare current image-generation options in a browser without building a Microsoft Foundry integration. Confirm the selected model shown in the interface, review the complete output, and use the platform's current pricing and terms. Model availability and controls can differ from Microsoft's direct products.

Official sources checked

Launch history comes from Microsoft's MAI-Image-1 announcement. MAI-Image-2.5 history and capabilities come from the MAI-Image-2.5 launch, MAI-Image-2.5 model card, and Microsoft Build 2026 announcement. Those sources were checked September 13, 2026. The separately cited 2.6 announcement and model card above were rechecked October 10, 2026.