Microsoft AI: MAI-Image-2.6

MAI-Image-2.6 is an image generation and editing model from Microsoft, released by Microsoft AI as the precision tier of the MAI-Image-2.6 family. It sits alongside a faster sibling, MAI-Image-2.6 Flash, which trades some fidelity for speed. The model is positioned for design-ready visuals, so it is aimed at work where the output needs to hold up in a finished layout rather than serve as a rough sketch.

On the input side, the model accepts text and images, and it produces images as output. That combination means it can generate from a prompt and also edit or transform images you supply as references. The context window is 4096 tokens, and the maximum output is 1024. Image behaviour is controlled through a small set of image parameters: an aspect ratio chosen from 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3 or auto; an n value fixed at exactly 1, so each request returns a single image; and input_references, which accepts between 0 and 5 reference images.

Those capabilities point to a fairly specific set of uses. Text-to-image generation covers concept art, marketing visuals, product mockups and illustration where you want a polished result from a written brief. Because the model also takes image input, it fits editing and variation work: supplying up to five references lets you guide style, composition or subject consistency across a series of outputs. The aspect ratio options make it straightforward to target a specific placement, such as a 16:9 banner, a 9:16 story format or a 1:1 social tile, without post-cropping.

You call the model through api14's OpenAI-compatible API using the model id "microsoft/mai-image-2.6". The supported request parameters are max_completion_tokens, max_tokens and temperature, alongside the image parameters described above for aspect ratio, n and input_references. Because the interface follows the OpenAI schema, existing clients and SDKs generally work with only the model id and endpoint changed.

A few constraints are worth weighing before you commit. The context window of 4096 tokens is modest, so very long or heavily detailed prompts may need trimming. Each request returns one image, so batch generation means multiple calls. If latency matters more than maximum fidelity, the Flash variant in the same family is the more natural fit; MAI-Image-2.6 is the choice when output quality is the priority.