Microsoft has launched MAI-Image-1[1], its first image-generation model developed entirely in-house, which represents a shift away from its dependence on OpenAI technologies, embedded deeply into most of its AI offerings, including Copilot.
The new model, currently being tested on the public image comparison platform LMArena, is expected to be added soon to Microsoft Copilot and Bing Image Creator.
The company says MAI-Image-1 is capable of generating photorealistic images, with improved rendering of natural lighting, reflections, and landscapes. Microsoft also claims that the model performs efficiently, delivering faster outputs with more consistency compared to some larger competitors, without naming any names. These capabilities were developed using feedback from professional artists and creative workers to avoid repetitive or low-quality results.
MAI-Image-1 has already reached the top 10 on LMArena, where users compare outputs from various text-to-image models in blind tests. The company describes this performance as an early indication of the model’s potential.
This release follows Microsoft’s introduction of two earlier models, MAI-Voice-1 and MAI-1-preview, launched over the summer.
In a previous interview, Mustafa Suleyman, who leads Microsoft’s AI division, said the company has a five-year development plan for its AI strategy and is committed to expanding its internal capabilities quarter after quarter.
According to Microsoft, MAI-Image-1 was trained using a carefully curated dataset, and its development prioritized safety, efficiency, and responsible output.
References
- ^ MAI-Image-1 (microsoft.ai)