Microsoft AI has announced MAI-Image-1, its first text-to-image generator developed entirely within the company. The model has already secured a position in the top 10 on LMArena, a public benchmark platform where human users compare and vote on the quality of outputs from different AI systems.
According to the official post, MAI-Image-1 is designed to produce diverse and non-generic results. Microsoft claims the model excels at generating photorealistic imagery, with a focus on details like lighting, reflections, and landscapes. The company also highlights its combination of speed and quality, stating it can process requests faster than many larger models.
To achieve this, Microsoft says it prioritized rigorous data selection and incorporated feedback from professionals in creative industries. The development of MAI-Image-1 is part of the company’s growing initiative to build its own AI models. This occurs alongside Microsoft’s major partnership with OpenAI and its recent use of models from Anthropic.
The company stated it is testing the model on LMArena to gather public feedback and ensure responsible outcomes. While MAI-Image-1 is currently only available on the benchmarking site, Microsoft announced it will be integrated into Copilot and Bing Image Creator soon.
Additional source: The Verge