MAI-Image-2.6 (Microsoft) for character sheets and covers
Every tier at a glance: which image tier for your sheets.
What MAI-Image-2.6 is
Microsoft AI announced MAI-Image-2.6 on August 10, 2026, and it became available to developers on Microsoft Foundry on September 4, 2026. Microsoft describes it as a diffusion model for both text-to-image and controlled image editing. At launch it placed second on the Arena text-to-image leaderboard, with the largest gains over its predecessor reported in portraits, text rendering and 3D-style images. Sources are at the end of this page.
On SceneMixer you do not need an Azure or Microsoft account; the tier is billed in SceneMixer credits.
What it draws here
The same things every image tier draws: each character full length on white, each location as an empty plate, each prop alone, the state images for second looks, and the three candidates of a promo cover. The video tiers then render every shot against those images.
Where to switch it on
Settings → Model Preferences → Image Model, then pick MAI-Image-2.6. From then on every new image is drawn on it: the character, location and prop sheets in step 2, regenerations, state images, and the promo cover posters. Images already drawn keep their pixels; switching never redraws anything. You can switch per group — leads on one tier, background places on another — and switch back.
About one megapixel
MAI-Image-2.6 has no resolution setting; on this site it draws at about one megapixel (768 by 1360 for a vertical image). The other five new tiers draw at about 2K. For a reference sheet that matters less than it sounds — the video model reads a face, a costume and a room from it — but it does matter if you plan to crop into a sheet or print a cover large. Download the cover at the size you need and check it before you use it in print.
A strict content filter
Every description is checked by the site first; a flagged description is not charged. MAI-Image-2.6 then runs Microsoft’s own filter, and in our use it refuses more readily than the other tiers. A refusal is refunded. If a harmless description is refused, reword it — plain, concrete sentences, no repeated phrases — or switch to another tier for that one sheet.
When to use it
- Portraits of the leads, if you like the look it gives your faces. Draw one lead on it and compare before moving the whole cast.
- Covers where the title is short and bold.
- Not for 2K location plates (SeedDream 5.0 Lite or GPT Image 2.5 Flare at 25 credits), or for a big cast on a budget (the default Qwen-Image 3.0 at 7 credits, with the free sheets).
What it costs
This tier is not covered by the free sheets. Every account’s 15 free sheets are drawn on the default image model, Qwen-Image 3.0, only. On MAI-Image-2.6 each sheet costs 31 credits from the first one (about $0.19 at the lowest credit price), and the price is printed on the button before you generate. Switch back to the default and whatever free sheets you have left are still there. A generation that is refused is not charged, and most failures are refunded automatically; a few kinds are held until support checks them. Regenerating a sheet is a new charge; the old image stays until you replace it.
Common errors
What you may meet on this tier:
| What you see | Usual cause | What to do |
|---|---|---|
| “Blocked by content safety rules — edit the script or description and retry.” | A word or phrase in the description tripped the site’s check, or the model maker’s own filter refused the picture | Edit the description and generate again; nothing was charged |
| “The generation service hit a temporary problem. You weren’t charged — feel free to try again.” | The model was saturated; the site kept the job queued for a while before giving up | Generate again in a few minutes, or switch tier |
| A sheet that ignores part of the description | Too many details competing in one description | Put identity details first; trim decoration |
| Generate All stops partway | Balance ran out; the remaining sheets were not charged | Top up, then generate the rest |
| Credits held for review | A rare failure where the model may already have been paid for the attempt | Nothing to do; support checks it and releases or refunds — see credits held for review |
All the image tiers, side by side
| Image tier | Credits per image | In short |
|---|---|---|
| Qwen-Image 3.0 (default) | 7 | The free sheets; a whole cast cheaply |
| Qwen-Image 3.0 Pro (Recommended) | 18 | Leads in close-up; Chinese descriptions |
| SeedDream 5.0 Lite | 8 | 2K locations and props on a budget |
| SeedDream 5.0 Pro | 25 | 2K, photographic skin, long descriptions read to the end |
| Nano Banana 2.1 | 19 | About 2K; Google’s faster model, the cheapest of the six new tiers |
| Nano Banana Pro | 46 | About 2K; detail kept from long descriptions; the most expensive tier |
| GPT Image 2.5 Flare | 25 | About 2K; OpenAI’s faster general-purpose model |
| GPT Image 2.5 Sunburst | 25 | About 2K; OpenAI’s slower, more precise model |
| Grok Imagine Image 2.0 | 35 | About 2K; xAI’s model, strong on lettering |
| MAI-Image-2.6 | 31 | About 1 megapixel; Microsoft’s model, strict content filter |
Sources
- Microsoft Foundry: MAI-Image-2.6 — model card; checked October 25, 2026.
- Windows Report: Microsoft releases MAI-Image-2.6 — launch and Arena placement; checked October 25, 2026.
MAI and Microsoft are trademarks of Microsoft, named here to identify the model offered as a tier. Prices are SceneMixer’s own, read from the live price list.
| MAI-Image-2.6 on SceneMixer | |
|---|---|
| Shown in Settings as | MAI-Image-2.6 |
| Used for | Character, location, prop and state sheets; cover posters |
| Credits per image | 31 |
| Output size | About 1 megapixel (no resolution setting) |
| Content filter | Microsoft’s own, strict; refusals are refunded |
| Free sheets | No — default model only |
FAQ
Can I use MAI-Image-2.6 on SceneMixer?
Yes. Pick MAI-Image-2.6 in Settings, Model Preferences, Image Model, and new sheets and covers are drawn on it.
How much does MAI-Image-2.6 cost here?
31 credits per image, about $0.19 at the lowest credit price.
Is it free on SceneMixer?
No. The 15 free sheets per account are drawn on the default image model only.
What resolution does MAI-Image-2.6 draw at?
About one megapixel on this site, for example 768 by 1360 for a vertical image. The model has no resolution setting.
Why was my description refused on MAI-Image-2.6?
Microsoft's own content filter runs on this tier and is strict. The refusal is not charged; reword the description or use another tier for that sheet.
Do I need an Azure account?
No. The tier is part of SceneMixer and billed in SceneMixer credits.
Is MAI-Image-2.6 Flash available?
No. The tier here is MAI-Image-2.6.
Is it good at faces?
Coverage of the model reports its biggest gains in portraits. Try it on one lead and compare with the tier you use now before switching the whole cast.
Try MAI-Image-2.6 on one lead
Generate one face on it, compare with your current tier, and keep whichever reads as your character.
Updated October 25, 2026