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

MAI-Image-1 is a legacy Microsoft image-generation model, not the current MAI image choice for a new workflow. Microsoft introduced it in October 2025 and added it to Bing Image Creator and Copilot Labs that November. In June 2026, Microsoft launched MAI-Image-2.5 and MAI-Image-2.5-Flash for text-to-image and image editing. Start with the current family unless you must reproduce an older MAI-Image-1 result.

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: MAI-Image-2.5 and MAI-Image-2.5-Flash add current text-to-image and image-to-image editing workflows.
• New-work recommendation: use a currently offered MAI-Image-2.5 route when its generation or editing capabilities fit; keep MAI-Image-1 only for legacy reproduction or research.

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. The current 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 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.

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 is the current family described by Microsoft.
• 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. Current status and capabilities come from the MAI-Image-2.5 launch, MAI-Image-2.5 model card, and Microsoft Build 2026 announcement. Sources were checked September 13, 2026.