After years of relying on OpenAI's artificial intelligence (AI) models to integrate into its products and services, Microsoft is finally starting to move into the independent development phase. Except for the Phi family of small language models , Microsoft has hardly ever announced a large AI model created by the company before. However, in August this year, the technology giant surprised everyone by introducing its first two internally developed models: MAI-Voice-1 and MAI-1-preview .
Today, Microsoft continues to expand its AI capabilities with the announcement of MAI-Image-1 – its first homegrown image generation model. This model is considered a direct competitor to gpt-image-1 (OpenAI) and Gemini/Imagen (Google).
According to Microsoft, MAI-Image-1 was trained with the goal of providing 'real value' to content creators. The team focused on preventing the model from generating repetitive or overly styled images. This was achieved through careful selection of training data and a sophisticated, task-specific evaluation process. Microsoft says the new model is capable of generating more realistic images, especially with landscapes and natural simulations.
In addition, Microsoft also asserts that MAI-Image-1 has superior processing speed compared to many other large models on the market, helping users create and refine images significantly faster.
In the Image Arena rankings of the community review platform LMArena, MAI-Image-1 debuted at number 9. Although this ranking is not high, it is an encouraging achievement for a first-generation model developed by Microsoft itself. Currently, competitors such as ByteDance, OpenAI, Google and Tencent are still holding the leading positions.
Mr. Mustafa Suleyman, CEO of Microsoft's AI division, shared on X (formerly Twitter) that the development team will continue to improve the model to improve the ranking in the future.
With the emergence of MAI-Image-1, the creative community and the tech world are watching to see if Microsoft can really challenge the current 'big guys' in the field of image-generating AI.