MICROSOFT AI · AI IMAGE GENERATOR
MAI-Image-2.6 is Microsoft AI's in-house image model. Write a scene with the exact words you want on signs or posters, turn on web search for real places and events, or edit with up to 5 reference images. Flash is the faster, cheaper version.
Review results, manage assets and access generation history in AI Studio.
Every example below was actually generated with MAI-Image-2.6 on ImgAI, with the original prompt and settings. Click "Use this example" to load it into the creator.
MAI-Image-2.6 · 1K · 3:4
The heading "TODAY'S BREWS" and all three menu lines with prices are spelled correctly on a realistic chalkboard under pendant lights.
A chalkboard menu on a cozy café wall. Hand-lettered heading "TODAY'S BREWS", then three lines: "Flat White $4.50", "Iced Matcha Latte $5.00", "Cardamom Bun $3.80". Warm pendant lights, a potted fern beside it, realistic photo.
MAI-Image-2.6 · 1K · 2:3
A smiling baker in a flour-dusted apron holds a round sourdough loaf; window light, shallow depth of field and a documentary feel.
Natural-light portrait of a smiling baker in a flour-dusted linen apron, holding a round loaf of sourdough in a small bakery, soft morning light from a window, shallow depth of field, documentary photography.
MAI-Image-2.6 · 1K · 2:3
Made with web search on: the Opera House and Harbour Bridge in flat screen-print colors under the "SYDNEY" title. The model also added a tagline at the bottom that the prompt did not ask for.
A retro travel poster of the Sydney Opera House at sunset seen from the harbour, with the big title "SYDNEY" at the top, flat screen-print colors.
MAI-Image-2.6 · 1K · 1:1
The mint robot turned coral red and got a small yellow umbrella; its face, parcel, pose and the street behind it match the original.
Change the robot's body color from mint green to coral red and give it a small yellow umbrella in one hand. Keep its face, pose, the parcel and the street background the same.
MAI-Image-2.6 Flash · 1K · 1:1
Made with Flash: a white leather sneaker with a terracotta sole floats above a soft shadow on a light grey background.
Clean e-commerce photo of a white leather sneaker with a terracotta sole, floating at a slight angle above a soft shadow, light grey seamless background.
MAI-Image-2.6 Flash · 2K · 1:1
Flash at 2K: desk, laptop, monstera, bookshelf and a round window with a night skyline in soft clay style. The small motivational signs were added by the model.
Isometric 3D render of a tiny home office: wooden desk, laptop, monstera plant, bookshelf and a round window with city lights at night, soft clay material, pastel colors.
Microsoft AI announced MAI-Image-2.6 in August 2026 as its strongest image model so far; on Sep 4, 2026 it said the model ranked No. 2 for text-to-image on Arena. Microsoft points to stronger text rendering, portraits and 3D imagery, plus web grounding and editing that combines people, products and scenes from several images. MAI-Image-2.6-Flash is the version for speed and volume.
On ImgAI you choose 1K or 2K (up to 1536×1536, which Microsoft calls 1.5K) and 13 aspect ratios from 1:4 to 4:1. Web search adds credits because the provider bills the looked-up content. When you edit with reference images, the model decides the output size itself — about 1 megapixel, close to the reference's shape. An image costs 25–105 credits without web search.
Based on the model's published capabilities and what we actually generated, to help you judge whether it fits your task.
Short lines of text rendered legibly in a realistic scene.
Natural light, skin and fabric with a documentary feel.
Turn on web search so landmarks and current subjects look right.
Clean e-commerce shots and isometric scenes, fast with Flash.
The tiers in this series share one page — switch between them in the creator's model menu.
The full model: best quality and text rendering. Use it for final images.
Faster and about half the price, with the same features. Good for drafts and volume.
Specs come from the configuration currently available on this site; the creator always shows the live price. Failed generations are refunded automatically.
| MAI-Image-2.6 | MAI-Image-2.6 Flash | |
|---|---|---|
| Input | Text + up to 5 reference images | Text + up to 5 reference images |
| Resolution | 1K · 2K | 1K · 2K |
| Prompt length limit | 4,000 characters | 4,000 characters |
Supported aspect ratios
| Tier | 1K | 2K |
|---|---|---|
| MAI-Image-2.6 | 50 | 105 |
| MAI-Image-2.6 Flash | 25 | 55 |
These come from real generations — read them alongside the original prompts in the examples.
Write the words you want, in quotes, and say where they go: heading "TODAY'S BREWS" at the top.
Web search suits real landmarks, people in the news and current events. For imagined scenes it only adds time and credits.
With reference images the size setting has no effect, so pick 1K to avoid paying for 2K.
Try ideas with Flash at about half the price, then rerun the keeper with the full model.
It is Microsoft AI's in-house image generation and editing model. ImgAI offers MAI-Image-2.6 and the faster MAI-Image-2.6-Flash.
25–105 credits per image without web search, depending on version and size; very wide or tall ratios at 1K cost a little more. Web search adds credits. Failed generations are refunded automatically.
The provider charges for the content it reads before drawing. In our tests the cost changed from run to run, so ImgAI adds a fixed amount when you turn it on.
Up to 5. Reference edits output about 1 megapixel at roughly the reference's shape, whatever size you pick.
Microsoft recommends the full model for final assets, complex edits and text-heavy designs, and Flash for fast, high-volume work. Flash costs about half.
The 2K option outputs up to about 2.36 megapixels (1536×1536 at 1:1). Microsoft calls this 1.5K; there is no 4K for this model.
Sign up and generate with credits. Results are saved to your library, ready to download or keep editing.