MAI Image 2.7 API — Microsoft's image model for generation and editing, with transparent backgrounds
mai-image-2.7MAI Image 2.7 is Microsoft AI's image model for text-to-image and editing with up to 5 input images, with transparent PNG / WebP and sizes up to 2048×2048; Microsoft reports it #1 on Artificial Analysis image editing. It is a public preview, returns one image per request and has no 4K. Call mai-image-2.7, billed per token at about $0.05 per 1K image; failed requests are not charged.
Generate Image
Result
Generated image will appear here
Enter a prompt and click Generate
#1 on Artificial Analysis image editing, per Microsoft
1,215 Elo over 5,478 votes as of 16 November 2026, 4 points ahead of GPT Image 2.6 Sunburst at max — close enough to call it a tie at the top. Microsoft also stresses consistency across successive edits.
Generate and edit with one model id
A prompt alone for text-to-image, or add 1 to 5 input images to recolour a product, swap a material, move an object or combine pictures.
Transparent backgrounds
background: transparent with PNG or WebP output returns cut-out product shots, logos, icons and stickers without a manual background-removal step.
1K or 2K, up to 2048×2048
Seven aspect ratios (1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16) at 1K or 2K, or an exact size: each side at least 768 px, up to 4,194,304 pixels, rounded down to a multiple of 16. No 4K.
Billed per token, about $0.05 per 1K image
Text input $7.15, image input $11.44, image output $48.62 per 1M tokens. Typical: 1K about $0.05, 2K about $0.20, a 1K edit with one input image about $0.06.
Preview model, not 100%
Microsoft lists it as a public preview. It can invent specifications or labels, so check text and numbers. Not every request succeeds; failed requests are not charged.
MAI Image 2.7 API: Microsoft's image model for generation and editing, with transparent backgrounds
MAI Image 2.7 is Microsoft AI's image model, released at Microsoft Ignite on 17 November 2026 as a public preview. Send a prompt for text-to-image, or add 1 to 5 input images to change an object's colour, material or position, restyle a photo or combine pictures. It can return a transparent background, and sizes go up to 2048×2048. Microsoft reports it #1 on the Artificial Analysis image-editing leaderboard.
On APIMODELS it is one model id, mai-image-2.7, on POST /v1/images/generations, billed per token on what each request uses: image output $48.62 per 1M tokens, about $0.05 for a typical 1K image. It returns one image per request, has no 4K, mask or seed, and not every request succeeds; the limits are listed below.
- Model ID
mai-image-2.7- Input
- Text prompt, plus 1–5 input images to edit (optional)
- Output
- 1 image, 1K or 2K, up to 2048×2048; PNG / JPEG / WebP; transparent background
- Aspect ratio
- 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16 or an exact size
- Price
- Per token; about $0.05 per 1K image
MAI Image 2.7 API limits at a glance
What the model accepts and what it does not, before you build on it.
| Supported | Not supported | |
|---|---|---|
| Input | A text prompt of up to 32,000 characters, alone (text-to-image) or with 1 to 5 input images to edit, restyle or combine | Masks (inpainting); more than 5 input images |
| Images per request | One | Several images in one call (send separate requests) |
| Size | aspect_ratio 1:1, 4:3, 3:4, 3:2, 2:3, 16:9 or 9:16 with resolution 1K (about 1 MP, the default) or 2K (each side doubled); or an exact WIDTHxHEIGHT with each side at least 768 px and at most 4,194,304 pixels in total (2048×2048), rounded down to a multiple of 16 | 4K; sides under 768 px; other aspect ratios such as 21:9 |
| Edits without a size | The model picks a size close to the input image | |
| Background | Opaque (default) or transparent; transparent needs PNG or WebP output | Transparent JPEG (JPEG has no alpha channel) |
| Output format | PNG (default), JPEG or WebP; jpg is accepted as jpeg | GIF, SVG |
| Size priority | image_size (a preset name or {"width", "height"}) wins over size, which wins over aspect_ratio; resolution applies to presets only and is ignored for exact sizes | |
| Other controls | quality is accepted but ignored | seed, negative_prompt, mask and n above 1 return a 400 instead of being silently dropped (results are not repeatable) |
Not every MAI Image 2.7 request succeeds
A public preview model with the usual failure modes. You pay only for requests that deliver an image.
Microsoft lists the MAI Image 2.7 family as a public preview, so behaviour, output sizes and quality may change while it is in preview. A request can fail after it is accepted: content review can refuse the prompt or an input image, and a generation can hit a temporary error. A failed request returns the reason and costs nothing. We have not run enough requests to publish a success rate or typical timings, so we publish neither.
Check text before you ship it. A hands-on review published on 18 November 2026 found MAI Image 2.7 labels readable in a dense technical illustration, and warns that image models of this generation still invent specifications, dates and part names. Treat model numbers, prices, ingredients and dimensions in a generated image as drafts until a person has read them.
MAI Image 2.7 API pricing
Billed per token on what each request actually uses. The typical costs below are references, not fixed unit prices.
| mai-image-2.7 | mai-image-2.7-flash | |
|---|---|---|
| Text input, per 1M tokens | $7.15 | $2.50 |
| Image input, per 1M tokens | $11.44 | $3.58 |
| Image output, per 1M tokens | $48.62 | $14.30 |
| Typical 1K text-to-image (about 1,024 output tokens) | about $0.05 | about $0.015 |
| Typical 2K image (each side doubled) | about $0.20 | about $0.059 |
| Typical 1K edit with one input image | about $0.06 | about $0.018 |
| Failed request | Not charged | Not charged |
100 product shots at 1K
≈ $5
100 × about $0.05; failed ones are not charged
One 2K hero banner
≈ $0.20
about 4,096 output tokens × $48.62 per 1M
One 1K edit of a product photo
≈ $0.06
about $0.05 of output plus the input image tokens at $11.44 per 1M
What MAI Image 2.7 is good for, and what it is not
Local edits to product photos
Change an object's colour, material or position and keep the rest of the scene. Microsoft leads with this, and with consistency across successive edits.
Transparent cut-outs
background: "transparent" with PNG or WebP returns product shots, logos, icons and stickers ready to place on any layout.
Combining up to five pictures
Put the product from one image into the scene of another, or merge a logo, a model and a background in one request.
4K, animated GIF or SVG
The largest output is 2048×2048 and the formats are PNG, JPEG and WebP. For 4K, GIF or SVG, use GPT Image 2.6.
Text that must be right without a check
Labels come out readable, but specifications, dates and part names can be invented. Proofread before anything ships.
Masks, many outputs per call, repeatable seeds
There is no mask or seed and one image comes back per request. For up to 10 outputs per call and masks, use GPT Image 2.6.
MAI Image 2.7 or MAI Image 2.7 Flash?
Both take exactly the same request: the same fields, limits, sizes and transparent background. Only the model id, Microsoft's positioning and the per-token rates differ, so you can switch with one string and compare on your own prompts.
| mai-image-2.7 | mai-image-2.7-flash | |
|---|---|---|
| Microsoft's positioning | The flagship; #1 on Artificial Analysis image editing per Microsoft (1,215 Elo, 16 Nov 2026) | The tier for production workloads; runs Canvas in MAI Labs. No leaderboard figure published |
| Image output per 1M tokens | $48.62 | $14.30 (about 30% of 2.7) |
| Typical 1K image | about $0.05 | about $0.015 |
| Inputs, sizes, transparency | Same | Same |
| Use it for | Edits and final images that matter most | Volume: catalogue variants, drafts, in-app editing |
Call it in one request
POST /v1/images/generations with your apimodels.app API key. Leave out image / images for text-to-image; add them to edit. By default you get a taskId back and poll GET /v1/images/generations?task_id= (or pass a callback_url); with an exact WIDTHxHEIGHT size and no callback_url, the image comes back in the same response. Every field is in the docs.
# Edit: transparent cut-out of a product photo at 1K
curl -X POST https://api.apimodels.app/v1/images/generations \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "mai-image-2.7",
"prompt": "Cut out the ceramic mug on a transparent background. Keep the mug, its handle and its glaze exactly as they are.",
"image": "https://example.com/mug.jpg",
"background": "transparent",
"output_format": "png",
"resolution": "1K"
}'
# → { "code": 200, "data": { "taskId": "...", "state": "pending", "model": "mai-image-2.7" } }MAI Image 2.7 vs GPT Image 2.6 Sunburst
The two models Microsoft names in its own comparison, both callable on apimodels.app with the same API key. The leaderboard row is Microsoft's statement; the other rows are each model's request limits on apimodels.app.
| MAI Image 2.7 | GPT Image 2.6 Sunburst | |
|---|---|---|
| Artificial Analysis image editing (per Microsoft, 16 Nov 2026) | #1, 1,215 Elo over 5,478 votes | 4 Elo behind, at the max setting |
| Largest output | 2048×2048 (4,194,304 pixels) | 4K, up to 3840 px on the long edge |
| Input images per request | Up to 5 | Up to 16 |
| Images per request | 1 | Up to 10 |
| Transparent background | Yes (PNG / WebP) | Yes |
| Animated GIF, SVG, masks | No | Yes |
| Billing on apimodels.app | Per token | Per image for plain images; per token for GIF, masks and transparent backgrounds |
The Elo figures are published by Microsoft, not measured by apimodels.app. A 4-point gap on a vote-based leaderboard is small enough to read as a tie at the top. The figure is for MAI Image 2.7; Microsoft has not published one for Flash.
Sources
Where the third-party facts on this page come from (checked 18 November 2026). Prices and limits are apimodels.app's own.
- Microsoft AI: Introducing MAI Image 2.7 ↗
Release at Ignite, public preview, Artificial Analysis image-editing result, Flash positioning and Canvas, transparent backgrounds.
- CIAOPS: I ran MAI Image 2.7 through my usual espresso machine test ↗
Hands-on review: readable dense labels; image models still invent specifications, dates and part names.
About MAI Image 2.7
MAI Image 2.7 is an Image Generation API provided by Microsoft. MAI Image 2.7 is the image generation and editing model from Microsoft AI (MAI), released on 17 November 2026 at Microsoft Ignite and offered by Microsoft as a public preview. On apimodels.app it is one model id, mai-image-2.7, on POST /v1/images/generations: send a prompt alone for text-to-image, or add 1 to 5 input images to edit them — change an object's colour, material or position, restyle a photo or combine pictures — on the same request. Microsoft reports that MAI Image 2.7 ranks first on the Artificial Analysis image-editing leaderboard with 1,215 Elo over 5,478 votes as of 16 November 2026, 4 points ahead of GPT Image 2.6 Sunburst at its max setting. A 4-point gap is small enough to call it a tie at the top rather than a clear win. Microsoft also stresses consistency across successive edits, so an image can be refined over several rounds without starting over. It can return a transparent background (background: transparent, with PNG or WebP output) for product shots, logos, icons and stickers. Output is PNG by default, or JPEG / WebP. Sizes come from an aspect ratio — 1:1, 4:3, 3:4, 3:2, 2:3, 16:9 or 9:16 — at 1K (about 1 megapixel, the default) or 2K (each side doubled), or from an exact WIDTHxHEIGHT: each side at least 768 px, at most 4,194,304 pixels in total (2048×2048), rounded down to a multiple of 16. There is no 4K, and each request returns one image. Pricing is per token, on the tokens each request actually uses: text input $7.15, image input $11.44 and image output $48.62 per 1M tokens. A 1K image is about 1,024 output tokens, so as a reference a 1K text-to-image costs about $0.05, a 2K image about $0.20 and a 1K edit with one input image about $0.06. These are typical figures, not a fixed unit price. Failed requests are not charged. Know the limits before you build on it. It is a preview model, so its behaviour may change. It renders dense, readable text, but like other image models of this generation it can still invent specifications, dates and part names, so check every word and number before you publish. There is no mask, seed or negative prompt. Not every request succeeds: content review can refuse a prompt or an image, and temporary errors happen; a failed request costs nothing. mai-image-2.7-flash takes exactly the same request at about 30% of these per-token rates. Through APIMODELS platform, you can access this model via a unified API with transparent pay-as-you-go pricing. Current pricing: Per token · text input: $7.15 / 1M tokens, Per token · image input: $11.44 / 1M tokens, Per token · image output: $48.62 / 1M tokens, Typical · 1K image: ≈ $0.05, Typical · 2K image: ≈ $0.20, Typical · 1K edit, one input image: ≈ $0.06.
Related model: MAI Image 2.7 Flash — the same request at about 30% of the per-token rates — about $0.015 per 1K image
Related model: GPT Image 2.6 Sunburst — level with MAI Image 2.7 on Artificial Analysis image editing, per Microsoft; adds 4K, animated GIF, SVG, masks and up to 16 reference images
Related model: GPT Image 2.6 Flare — OpenAI's default GPT Image 2.6 tier — 4K, animated GIF and SVG, up to 10 images per call
Related model: Seedream 5.5 — pixel-level edits, native RGBA output and layer splitting into up to 100 transparent layers
Key Features
- #1 on Artificial Analysis image editing, per Microsoft -- 1,215 Elo over 5,478 votes as of 16 November 2026, 4 points ahead of GPT Image 2.6 Sunburst at max — close enough to call it a tie at the top. Microsoft also stresses consistency across successive edits.
- Generate and edit with one model id -- A prompt alone for text-to-image, or add 1 to 5 input images to recolour a product, swap a material, move an object or combine pictures.
- Transparent backgrounds -- background: transparent with PNG or WebP output returns cut-out product shots, logos, icons and stickers without a manual background-removal step.
- 1K or 2K, up to 2048×2048 -- Seven aspect ratios (1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16) at 1K or 2K, or an exact size: each side at least 768 px, up to 4,194,304 pixels, rounded down to a multiple of 16. No 4K.
- Billed per token, about $0.05 per 1K image -- Text input $7.15, image input $11.44, image output $48.62 per 1M tokens. Typical: 1K about $0.05, 2K about $0.20, a 1K edit with one input image about $0.06.
- Preview model, not 100% -- Microsoft lists it as a public preview. It can invent specifications or labels, so check text and numbers. Not every request succeeds; failed requests are not charged.
Use Cases
Product Photo Edits
Recolour a product, swap a material or move an object in an existing photo while the rest of the scene stays as it was — the editing strength Microsoft leads with. Send the photo as image and say what to keep.
Transparent Cut-Outs and Logos
Ask for background: transparent in PNG or WebP and place the result straight onto a store page, a banner or a slide, with no background-removal step.
Several Rounds of Revisions
Feed each result back as the input for the next change. Microsoft says MAI Image 2.7 keeps the image consistent across successive edits.
Posters, Labels and Diagrams with Text
Dense labels come out readable, but proofread names, dates and numbers: image models of this generation still invent specifications.
Combining Up to Five Pictures
Put the product from one image into the scene from another, or merge a logo, a model and a background in a single request.
Hero Images up to 2048×2048
2K output or an exact size up to about 4.2 megapixels for banners and page headers. There is no 4K; for that, use GPT Image 2.6.
Why APIMODELS
- Unified API -- One API key to access all models, no need to register on multiple platforms
- Transparent Pricing -- Pay-as-you-go with per-call pricing published on every model page, ideal for indie developers and startups
- Instant Access -- Start right after signup; pay by card, PayPal, Alipay or USDT
- Full Documentation -- Detailed API docs with code examples in cURL, Python, and Node.js
Frequently Asked Questions
What is the difference between APIMODELS and the official Microsoft API?
APIMODELS offers the same MAI Image 2.7 model through our aggregation platform. We provide a unified API interface so you do not need separate accounts for each provider - one API key to access all models.
How much does the MAI Image 2.7 API cost?
On apimodels.app mai-image-2.7 is billed per token, on the tokens each request actually uses: text input $7.15, image input $11.44 and image output $48.62 per 1M tokens. A 1K image is about 1,024 output tokens, so a 1K text-to-image costs about $0.05, a 2K image about $0.20 and a 1K edit with one input image about $0.06. Those are typical costs, not a fixed per-image price. Failed requests are not charged.
How do I call the MAI Image 2.7 API, and how do I edit an image with it?
POST https://api.apimodels.app/v1/images/generations with your API key and model mai-image-2.7. A prompt alone gives text-to-image. To edit, add the picture as image (or up to 5 pictures as images) and describe the change, for example "make the mug matte black and keep everything else the same". The call returns a taskId to poll, or send a callback_url; with an exact size such as 1024x1024 and no callback_url, the image comes back in the same response.
Is MAI Image 2.7 better than GPT Image 2.6?
On image editing the two are level. Microsoft reports MAI Image 2.7 at #1 on the Artificial Analysis image-editing leaderboard with 1,215 Elo over 5,478 votes as of 16 November 2026, 4 points ahead of GPT Image 2.6 Sunburst at its max setting, which is effectively a tie. The practical differences are features: GPT Image 2.6 on apimodels.app adds 4K, animated GIF, SVG, masks, up to 16 reference images and up to 10 outputs per call, while MAI Image 2.7 tops out at 2048×2048, takes up to 5 input images and returns one image per request. Both support transparent backgrounds. Run your own prompts on both before you switch.
Microsoft AI: Introducing MAI Image 2.7 ↗Does MAI Image 2.7 support transparent backgrounds?
Yes. Set background to transparent and output_format to png or webp; JPEG has no alpha channel, so a transparent request needs PNG or WebP. It suits product shots, logos, icons and stickers that you want to place on another layout without cutting them out by hand.
What sizes and resolutions does MAI Image 2.7 support?
Pick an aspect ratio — 1:1, 4:3, 3:4, 3:2, 2:3, 16:9 or 9:16 — with resolution 1K (about 1 megapixel, the default) or 2K (each side doubled), or pass an exact size as WIDTHxHEIGHT. Exact sizes need at least 768 px on each side and at most 4,194,304 pixels in total (2048×2048), and are rounded down to a multiple of 16. There is no 4K. When you edit without a size, the model picks one close to the input image.
Should I use MAI Image 2.7 or MAI Image 2.7 Flash?
Both take the same request: the same fields, sizes, input images and transparent background. Flash is billed at about 30% of the 2.7 per-token rates (image output $14.30 against $48.62 per 1M tokens), about $0.015 per 1K image against about $0.05, and Microsoft positions it for production workloads. The leaderboard result Microsoft published is for MAI Image 2.7. Use 2.7 for the edits and images that matter most and Flash for volume; switching is one string.
What are the limits of MAI Image 2.7, and what happens when a request fails?
Microsoft lists MAI Image 2.7 as a public preview, so its behaviour may change. One image per request, up to 5 input images, no mask, seed or negative prompt, and no 4K. It renders dense readable text, but it can still invent specifications, dates or part names, so proofread anything a customer will read. Not every request succeeds: content review can refuse a prompt or an image, and temporary errors happen. A failed request returns the reason and is not charged.
CIAOPS hands-on review, 18 Nov 2026 ↗How does MAI Image 2.7 compare with other models in the Image Generation category?
On APIMODELS, MAI Image 2.7 runs alongside 60+ models on one API key and one balance, so choosing is about fit, not lock-in. It supports Text to Image & Edit, Up to 5 Input Images, Transparent PNG / WebP, 1K / 2K up to 2048×2048, Per token · ≈$0.05 per 1K image, and you can weigh it on price and capability against other models in the Image Generation category, then switch by changing a single model-name string — no new account or integration. Browse every model in that category with live pricing at apimodels.app/models.
What can MAI Image 2.7 do?
MAI Image 2.7 supports: Text to Image & Edit, Up to 5 Input Images, Transparent PNG / WebP, 1K / 2K up to 2048×2048, Per token · ≈$0.05 per 1K image. See the APIMODELS docs for full parameters and call examples.
Can I use the MAI Image 2.7 API from any country, including mainland China?
Yes. APIMODELS does not restrict access by country: the site and the API answer requests from anywhere, and apimodels.app and api.apimodels.app are reachable directly from mainland China without a proxy. You call MAI Image 2.7 with one APIMODELS key, so you do not need a separate account with the model's developer.
What payment methods are supported? Can I pay with PayPal or crypto?
Yes to both. You can top up with a card (Visa, Mastercard and others, through Stripe), PayPal, Alipay, or USDT on TRON (TRC20) or BNB Smart Chain (BEP20). Card, PayPal and Alipay top-ups are credited immediately; USDT is credited once the transfer confirms on-chain. Auto top-up is optional, and a $50 top-up adds a $1 bonus ($3 on a $100 top-up). The balance is in USD and billed per use, with no subscription.