Drop files here, paste files, or add a URL. JPEG, PNG or WebP · up to 10 MB each
Example Output
This is sample data. Generate a new result to see real output.
Pay as you go. No subscription, no hidden fees.
| Model | Text input /1M tokens | Image input /1M tokens | Image output /1M tokens |
|---|---|---|---|
| MAI Image 2.6 | $4.00/1M tokens 800 credits/1M tokens $5.00/1M tokensSave 20% | $6.40/1M tokens 1280 credits/1M tokens $8.00/1M tokensSave 20% | $30.40/1M tokens 6080 credits/1M tokens $38.00/1M tokensSave 20% |
| MAI Image 2.6 Flash | $1.40/1M tokens 280 credits/1M tokens $1.75/1M tokensSave 20% | $2.00/1M tokens 400 credits/1M tokens $2.50/1M tokensSave 20% | $15.20/1M tokens 3040 credits/1M tokens $19.00/1M tokensSave 20% |
Charges are based on the final output.
Microsoft's image model for photorealistic generation, accurate in-image text, and multi-image editing
Access the MAI Image 2.6 API instantly on PoYo, no waitlist required. Microsoft built MAI-Image-2.6 for photorealistic images, lifelike people, product and commercial visuals, and text that reads correctly inside the image. One REST call covers text-to-image, single-image editing, and composition from up to 5 reference images.
Choose mai-image-2.6 for the highest quality or mai-image-2.6-flash for faster, lower-cost runs with the same request fields. Pick one of seven aspect ratios at 1K or 2K, or set a custom WIDTHxHEIGHT up to 1536×1536, and turn on enable_web_search when the image depends on current information.
Text-to-image, editing, and multi-image composition through one request format, with custom sizes and optional web search.
01
Stock photos rarely match a specific brief, and a staged shoot takes days to organize. Describe the subject, lens, light and motion, and MAI Image 2.6 returns an image that reads as a real photograph, with believable lighting, natural texture and camera effects such as motion blur and shallow depth of field. Microsoft lists photorealistic and cinematic quality, high-fidelity portraits and product imagery among the model's strongest areas.
02
Image models often misspell the words a poster or cover depends on. Text rendering is where Microsoft measured the largest gain over MAI Image 2.5, at +91 Elo on Arena. Put the exact copy in quotes and say where it goes, and MAI Image 2.6 sets headlines, subtitles and author names into the layout, so a design draft arrives with its text already in place.
03
Reshooting a product for every campaign costs time and money. Send one image in image_urls, describe the change and say what must stay the same. Microsoft built MAI Image 2.6 for precise edits that keep the composition intact, from new backgrounds and lighting to object replacement, attribute changes, text updates and motion-blur cleanup, so one product photo becomes many campaign shots.
04
A campaign visual usually starts from scattered material: a product shot, a color palette, a texture, a location and a mood reference. Pass up to 5 of them in image_urls and describe the result you want. In Microsoft's example, a color swatch, an action shot, a canyon rock texture, a knit close-up and a product photo become one canyon hero image of the same magenta running shoe from a single prompt.
05
Fixed square outputs rarely fit banners, stories and print layouts. Choose 1:1, 4:3, 3:4, 3:2, 2:3, 16:9 or 9:16 at 1K or 2K, or pass a custom WIDTHxHEIGHT. Microsoft limits the total pixel count rather than each side, so a wide banner such as 3072x768 works as long as it stays within 2,359,296 pixels.
06
Real landmarks, products and recent events need details a model may not have learned in training. Set enable_web_search to true and MAI Image 2.6 grounds the image in current web content before it generates. Microsoft's example asks for the Metrópolis Building in Madrid "as it stands today" at the corner of Calle de Alcalá and Gran Vía, and the result shows the building's dark dome, gilded details and street corner.
Send one product photo and generate lifestyle scenes, seasonal backgrounds, and room placements with MAI Image 2.6 while the product keeps its shape and color. Build a full catalog set without a new photo shoot.
Draft campaign visuals with the headline already in place, test layouts with MAI Image 2.6 Flash, and render the chosen concept with MAI Image 2.6. Custom sizes match each ad placement.
Create posts, stories, and covers in 1:1, 9:16, or 16:9 from short prompts. MAI Image 2.6 Flash keeps high-volume content queues fast and low-cost.
Combine a room photo with furniture or decor references to preview staging ideas. MAI Image 2.6 matches lighting and shadows so the result reads as one photograph.
Produce book covers, posters, and lesson illustrations with titles rendered inside the image. Use web search for current places and events, then review the details before publishing.
Add text-to-image and editing to your product or agent with one REST call. Submit a task, get a task_id at once, and receive the image by polling or webhook.
Microsoft AI's two MAI Image 2.6 models side by side, based on Microsoft's announcements and model card. Both accept the same request fields and support the same features on PoYo, so the choice comes down to maximum quality or faster, lower-cost generation.
| Feature | MAI Image 2.6 | MAI Image 2.6 Flash |
|---|---|---|
| Microsoft's positioning | Best overall: high-quality generation with precise, controllable editing | Fastest and lowest cost: production-ready quality at less than half the flagship's price |
| Key advantage | Highest overall quality: photorealism, portraits, product visuals and +91 Elo text rendering over MAI Image 2.5 | Speed and cost: 2.8x faster than GPT-Image-2-Medium with 72% greater efficiency, with the same features as 2.6 |
| Best for | Final campaign assets, complex edits, detailed commercial design and high-fidelity text | Interactive apps, automated pipelines, personalization and high-volume production |
| Leaderboard rank | No. 2 text-to-image and editing on Arena; No. 2 text-to-image and No. 1 editing on Artificial Analysis (as of September 4, 2026) | No separate rank published |
| Text-to-image | Yes | Yes |
| Image editing | Yes, up to 5 reference images | Yes, up to 5 reference images |
| Web grounding | Yes | Yes |
| Maximum output | 2,359,296 pixels (1536×1536 equivalent) | 2,359,296 pixels (1536×1536 equivalent) |
| Release | Announced August 10, 2026 | September 4, 2026 |
Step 1: Register & Create API Key
Sign in to PoYo and create an API key in the dashboard. The same key works for MAI Image 2.6, MAI Image 2.6 Flash, and every other model on PoYo.
Create API Key →
Step 2: Top Up Credits
Add credits to your account. Each request reserves a maximum based on its size, prompt, and reference images, then is charged by the tokens it actually used; failed tasks are not charged.
Add Credits →
Step 3: Integrate the API
Send POST https://api.poyo.ai/api/generate/submit with "model": "mai-image-2.6" or "mai-image-2.6-flash" and an input object with prompt, plus optional image_urls, size, resolution, and enable_web_search. Poll the returned task_id or pass a callback_url to receive the finished image.
View Documentation →
Call mai-image-2.6 and mai-image-2.6-flash with the same request format and API key you use for every other image, video, and audio model on PoYo.
See the token rates and the reserved maximum before you submit. The final charge follows actual usage, never exceeds the reserve, and failed tasks are free.
Try prompts, aspect ratios, custom sizes, reference images, and web search in the browser, then copy the JSON request into your code.
Use documented request fields, task polling, and callback URLs, so long renders never block your application.
