The competition in AI image generation and editing has reached a new milestone. On November 17, Microsoft officially launched a new generation of image generation and editing models—MAI Image2.7 and MAI Image2.7Flash—and simultaneously launched Canvas, an interactive creative space powered by Flash.
Precise local edits: Change only what needs changing while maintaining full visual consistency
The core highlights of this update focus on "precise retouching" and "visual stability." According to Microsoft, MAI Image2.7 can flexibly adjust the color, material, or spatial position of specific objects in an image while ensuring the rest of the image remains highly stable.
In scenarios involving successive edits, the model can perfectly maintain overall visual consistency. In addition, both models natively support transparent background output, which can be used directly to create product promotional images, logos, icons, and various types of illustrations, greatly reducing the tedious manual work of background removal afterward.

To give creators a smoother experience, Microsoft also launched Canvas, an interactive creation space. It integrates generation and editing features into a single workspace. Users can start creating from text prompts or hand-drawn sketches, or directly select a local area of an image to replace, move, or fine-tune its style.

Overwhelming value for money: top global performance at only one-fifth the cost of GPT Image2.6
In terms of ecosystem integration and pricing, both models are now officially in public preview on Microsoft Foundry. The officially announced costs per thousand images generated and edited are as follows:
MAI Image2.7: about $35 (approximately RMB 233.8 at the current exchange rate)
MAI Image2.7Flash: about $10 (approximately RMB 66.8 at the current exchange rate)
In authoritative industry evaluations, MAI Image2.7 has shown extremely impressive competitiveness. According to data as of November 16, the model topped the Artificial Analysis image editing leaderboard with a high score of 1215, beating GPT Image2.6Sunburst (max) by 4 points, while its operating cost is only about one-fifth of the latter's. Microsoft also noted that actual performance and cost comparisons may vary depending on model configuration, specific tasks, and evaluation methods.




