Microsoft's Chief AI Officer Mustafa Suleiman announced the launch of the second-generation image generation model MAI-Image-2. This model performed strongly in the authoritative LMArena evaluation ranking, directly rising to the third position globally.
In the LMArena ranking, known as the "highest competition arena" in the AI image generation field, MAI-Image-2 has attracted attention: it currently ranks just behind Google's Gemini-3.1-flash-image-preview and OpenAI's GPT-image-1.5-high-fidelity. Compared to the first-generation model released in October 2025 (initially ranked ninth), the second-generation model has achieved an essential breakthrough in overall quality.
Technical Highlights: Solving the Industry Pain Point of "Text Corruption"
MAI-Image-2 not only significantly improves visual effects but also solves the long-standing problem of text rendering in AI-generated images:
Precise Text Rendering: It significantly enhances the ability to handle information charts, presentation slides, and complex logic charts containing text, with clear and non-corrupted text.
Ultra-Realistic: It can accurately restore natural lighting, realistic skin tones, and build realistic environments that follow physical laws.
Movie-Level Composition: It supports generating ultra-high-resolution images with surreal concepts, elaborate compositions, and grand worldviews.
Microsoft is accelerating the delivery of this top-tier capability to users:
Immediate Experience: Users can now log in to the MAI Playground platform for free trial use.
Product Integration: MAI-Image-2 is gradually being integrated into Copilot and Bing Image Creator, allowing a large number of ordinary users to directly access it in daily work and creation in the future.
This release marks that Microsoft has firmly entered the first tier in the multimodal generation field. By solving this core pain point of text rendering, it further expands the application scenarios of AI image generation in professional office fields.
