aifollow.news Search
Back IT之家 科技新闻
IT之家 科技新闻· · Original publication time AI score60

Microsoft launches MAI-Image-2.7-Preview for image editingMachine translation

微软推出 MAI-Image-2.7-Preview,开放图像编辑体验

Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.

AI-assisted summary

Microsoft has launched the MAI-Image-2.7-Preview image editing model, which users can try on Arena, Artificial Analysis and MAI Playground. It currently edits existing images using text instructions; text-to-image generation is not yet available. Microsoft says it improved over MAI-Image-2.6 across all 10 use cases and seven editing categories in an Artificial Analysis evaluation.

Article · AI translation

IT Home reported on October 28 that Microsoft issued an announcement yesterday (October 27) introducing the MAI-Image-2.7-Preview model. Mustafa Suleyman, CEO of Microsoft AI, proudly declared: “MAI-Image is now the world's second-ranked image editing model!”

Suleyman wrote in a post that, according to Artificial Analysis's image editing rankings, Microsoft's MAI-Image series has once again outperformed Google, Meta, xAI and GPT-Image-2.5-Flare to become the world's second-ranked image editing model, behind only OpenAI's GPT Image 2.5 Sunburst (max). IT Home provides the relevant screenshot below:

It is worth noting that the MAI-Image-2.7-Preview model currently focuses on “original image input + text instructions → edited image output.” Text-to-image generation is not yet available. It is a controllable editing model for existing photos, design drafts, screenshots or film and television footage.

Microsoft emphasized that it can handle tasks that are “harder to get right”: changing an image's overall style, relighting it, modifying text or symbols in the image, restoring and enhancing source material, and complex edits that require an understanding of the relationships among objects, lighting and the environment.

Microsoft said that MAI-Image-2.7 has improved over its predecessor, MAI-Image-2.6, across all 10 use cases and 7 types of editing action evaluated by Artificial Analysis, and has reached the leading edge of its category in 4 areas: enhancement and restoration, reasoning-based editing, scene and style, and text and symbols.

Its lead is largest in scene and style editing, where it scores 24 Elo points higher than the next-best model. On Artificial Analysis's AA-Image-Editing v2.1 leaderboard, it ranks 2nd with 1,193 Elo, behind only OpenAI's GPT Image 2.5 Sunburst (max).

Users can currently try it on Arena, Artificial Analysis and MAI Playground. Only image editing is available for now; Microsoft says text-to-image generation is “coming soon” but has not announced a date.

The model is still in Preview. Microsoft has not announced a public API, pricing or a date for general commercial availability. The model weights have not been released either, so it cannot be self-hosted. Playground availability may also be limited by region.

Found an error? Send a correction