Small output
About 1 megapixel per image, so keep it for screens rather than print.
Image model by Microsoft AI
MAI-Image-2.5
Microsoft's image model is at its best on the words that sell a product: labels, packaging and signs, spelled the way you wrote them.
Make an image with MAI-Image-2.5
The prompt
Retail product photograph of three small-batch hot sauce bottles from an invented brand, "EMBER & ASH", standing in a row on a weathered reclaimed-wood shelf against a dark charcoal plaster wall. Each is a slim 150 ml glass woozy bottle with a black wax-dipped cap and a matte kraft-paper label with a hand-drawn flame illustration. Labels, left to right, clearly legible: "EMBER & ASH 路 SMOKED HABANERO 路 HEAT 4/5", "EMBER & ASH 路 GREEN JALAPE脩O & LIME 路 HEAT 2/5", "EMBER & ASH 路 GHOST PEPPER GOLD 路 HEAT 5/5". The sauces glow deep red, bright green and amber through the glass. Dried chilies and a sprig of cilantro scattered on the shelf, warm side light, crisp label detail, premium craft-food styling.
Made with MAI-Image-2.5 on Predis.ai on October 27, 2026, 16:9. Three labels, every word as written, the 帽 included.
Seven shapes
Seven shapes, one bottle
The same prompt in every shape Predis.ai offers for this model, from a square feed post to a 9:16 story. The bottle, the label and the lime stay recognizably the same shoot.
The prompt, used seven times
Product photograph of a single bottle of "EMBER & ASH" Smoked Habanero hot sauce: a slim glass woozy bottle with a black wax-dipped cap and a matte kraft-paper label with a hand-drawn flame, reading "EMBER & ASH" and "SMOKED HABANERO" and "HEAT 4/5". It stands on a charred oak board beside a few dried red chilies, half a lime and a pinch of coarse sea salt, dramatic warm side light against a dark background, crisp label detail.
Made with MAI-Image-2.5 on Predis.ai on October 27, 2026, one run per shape.
Same truck, two looks
One brief written twice, once as a night photograph and once as flat sticker art. The logo, the lettering and the menu board carry across both.
Both prompts
Photo: Photograph of a matte-black food truck for an invented hot sauce brand, "EMBER & ASH", parked at a busy night market under strings of warm bulbs. The truck's side is painted with a big hand-drawn orange flame logo and the words "EMBER & ASH 路 TACOS & HOT SAUCE" in bold cream lettering; a serving window glows warm with a cook passing out tacos, a chalkboard menu reads "Smoked Brisket Taco 6", "Charred Corn 4", "Ghost Gold Wings 9". A small crowd queues, smoke curls from a grill. Cinematic photorealism, rich warm night light, crisp sign lettering. Cartoon: A bold flat cartoon illustration of the same scene: the matte-black EMBER & ASH food truck at a night market, the orange flame logo and "EMBER & ASH 路 TACOS & HOT SAUCE" lettering, a smiling cook leaning out of the glowing window holding a taco, the same chalkboard menu, a few cheerful customers queueing. Thick clean outlines, flat colors, a limited palette of charcoal, flame orange, cream and teal, playful sticker-art style, crisp lettering.
Made with MAI-Image-2.5 on Predis.ai on October 27, 2026, 3:2.
About 1 megapixel, every time
Every image holds roughly the pixels of a 1024 脳 1024 square, whatever its shape. That's sharp on a phone and on the web. For a billboard or a printed catalog, pick a model with bigger output, such as Nano Banana Pro or GPT Image 2.5.
Made byMicrosoft AI
ReleasedJune 2, 2026
Starts fromA written prompt
Shapes7, from 1:1 to 9:16
Per runUp to 4 images
Before you build a campaign on it
Words only on Predis.ai
There's no reference photo here, so describe your product and label in full, word for word.
No real people
Microsoft keeps real people and public figures out of scope. Use invented characters.
Confident details
It can invent plausible details that aren't true, such as a menu price or a spec. Check the small print against your real product.
MAI-Image-2.5 questions
What is MAI-Image-2.5?
It's the image model from Microsoft AI, Microsoft's in-house AI group, released on June 2, 2026. Microsoft says its strongest area is text inside images, such as labels, packaging, posters and signs, followed by product and brand imagery.
Is MAI-Image-2.5 Microsoft's newest image model?
No. Microsoft has since released MAI-Image-2.6. Predis.ai offers MAI-Image-2.5, which you can also find in Microsoft products such as PowerPoint and OneDrive Photos.
What size are MAI-Image-2.5 images?
About 1 megapixel, the same number of pixels as a 1024 脳 1024 square. A landscape 16:9 image comes out at 1360 脳 768. That's plenty for social posts and web pages, but too small for large print.
Which shapes does it make on Predis.ai?
Seven: 1:1, 4:3, 3:4, 16:9, 9:16, 3:2 and 2:3. You can make up to 4 images from one prompt.
Can I edit my own photos with MAI-Image-2.5 on Predis.ai?
Not on Predis.ai, where it works from a written prompt only. To change a photo you already have, use an editing model such as Nano Banana Pro or GPT Image 2.5.
Can it make images of real people?
No. Microsoft keeps real people and public figures out of scope, so stick to invented characters, products and places.
Put your product on the shelf
Describe the label word for word and MAI-Image-2.5 prints it.