Microsoft AI's in-house image model, ranked No. 3 for text to image on Arena at launch. Photorealistic lighting, clean product imagery and readable text, 2 credits per image, five aspect ratios, no watermark. Available on every Aura AI plan.
MAI Image 2.5, written MAI-Image-2.5 by Microsoft, is the third generation of the image models that Microsoft AI trains in-house rather than licensing from OpenAI. The line started with MAI-Image-1, announced on October 13, 2025, which debuted at No. 9 on the LMArena text to image leaderboard and became a selectable model in Bing Image Creator next to DALL-E 3 and GPT-4o. MAI-Image-2 followed in the MAI Playground in March 2026 and in Microsoft Foundry on April 2, 2026. MAI-Image-2.5 was released on June 2, 2026 together with a faster Flash variant.
At launch, Microsoft reported that MAI-Image-2.5 ranked No. 3 for text to image and No. 2 for image editing on the Arena leaderboard, with an overall gain of 75 points over MAI-Image-2. The largest improvements were in text rendering (107 points) and in the cartoon, anime and fantasy category (90 points). Microsoft describes the model's strengths as detailed, coherent images with stronger product imagery and prompt adherence, an understanding of scene structure, lighting and spatial relationships, and edits that preserve facial identity across changes of pose or viewpoint. It powers image generation in PowerPoint and editing in OneDrive, and is available to developers through Foundry, the MAI Playground and OpenRouter.
On Aura AI, MAI Image 2.5 is available in the Text to Image tool as MAI Image 2.5 and MAI Image 2.5 Pro, both at 2 credits per image, in five aspect ratios and without watermark. There is no edit variant yet, so pair it with FLUX Kontext or Nano Banana Edit when you need to change a result. It sits next to FLUX 2 and Gemini 3 Image in the same selector.
Six reasons to pick Microsoft's model for a given image
Microsoft positions the model around scene structure, lighting and spatial relationships. Interiors, portraits and outdoor scenes come out with believable shadows, reflections and depth rather than a flat, over-processed look.
The biggest measured jump over MAI Image 2 is in text rendering. Labels, headlines and packaging copy inside the picture come out legible, which matters for ads and mockups.
Stronger product imagery is one of the three improvements Microsoft lists for 2.5. Bottles, devices and packaging keep their proportions and material finish, from matte plastic to brushed metal.
Multi-part prompts with a subject, a setting and a constraint are followed more reliably than in earlier MAI releases, so fewer re-rolls are needed for specific briefs.
The second largest gain on Arena, 90 points, is in stylised work. Character sheets, illustrated covers and fantasy scenes benefit as much as photorealistic shots.
On Aura AI both MAI Image 2.5 and MAI Image 2.5 Pro cost 2 credits per image, one less than Nano Banana Pro and two less than GPT Image 2, at any of the five aspect ratios.
Four steps from prompt to a finished image
Flat cost per image, identical for every aspect ratio
| Variant | Mode | Credits per image | Best for |
|---|---|---|---|
| MAI Image 2.5 | Text to image | 2 credits | Everyday generation, social visuals, drafts |
| MAI Image 2.5 Pro | Text to image | 2 credits | Final renders, product shots, images with text |
How Microsoft's model compares to its siblings on Aura AI
| Feature | MAI Image 2.5 | Nano Banana | FLUX 2 | Gemini 3 Image |
|---|---|---|---|---|
| Maker | Microsoft AI | Black Forest Labs | ||
| Text to image credits | 2 | 3 | 1 to 3 | 1 to 2 |
| Image edit on Aura AI | No | ✓ 2 to 3 credits | ✓ 2 to 3 credits | ✓ 1 to 2 credits |
| Variants on Aura AI | 2 | 2 | 4 | 2 |
| Released | June 2, 2026 | November 20, 2025 (Pro) | November 25, 2025 | February 26, 2026 (3.1 Flash) |
| Text rendering | ✓ +107 Arena points vs 2.0 | ✓ Yes | ✓ Yes | ✓ Yes |
| Aspect ratios on Aura AI | 5 | 5 | 5 | 5 |
| Plan | All plans | All plans | All plans | All plans |
Where photorealistic lighting and clean text pay off
Studio-style packshots and lifestyle placements with accurate materials and shadows. Generate the hero image in 1:1 for the listing and 9:16 for the story format from the same prompt.
Social ads where the copy is part of the picture. The text rendering gains in 2.5 make headlines and price tags readable at feed size.
The same model that generates images inside PowerPoint is available here for slides, reports and internal decks, with consistent style across a full set of pictures.
Anime and fantasy work improved by 90 points on Arena in this release. Use it for character concepts, book covers and game art before moving to an edit model for variations.
Common questions about MAI Image 2.5 on Aura AI
Microsoft's photorealistic image model at 2 credits per image, five aspect ratios, no watermark. Available on every plan, worldwide.