Test Microsoft’s New AI Image Generator Now – Here’s How
▼ Summary
– Microsoft is testing its own in-house AI image generation model called MAI-Image-1, which aims to produce less repetitive and more realistic images.
– The model is currently available for public testing and ranking on the LMArena leaderboard site, where it has debuted in the top ten.
– MAI-Image-1 will soon be integrated into Microsoft’s Copilot and Bing Image Creator, replacing the current OpenAI models like DALL-E-3.
– Microsoft has developed other in-house AI models, including MAI-Voice-1 for speech and MAI-1-preview for text, also testable on LMArena.
– This shift reflects Microsoft’s move to bring more AI development in-house as its partnership with OpenAI has recently weakened.
Microsoft has launched its own internal AI image generation model, allowing users to test it directly through the LMArena leaderboard platform. This new model, named MAI-Image-1, is designed to produce photorealistic images with improved lighting, reflections, and natural landscapes, offering both speed and high-quality output. While it will eventually be integrated into Copilot and Bing Image Creator, for now it can be accessed and evaluated by the public on LMArena, where it has already secured a spot in the top ten rankings.
Currently, Microsoft relies on OpenAI’s GPT-4o and DALL-E 3 for generating images in Copilot and Bing Image Creator. However, the company is now shifting more AI development in-house, and MAI-Image-1 represents a significant step in that direction. According to Microsoft, the model was trained with input from creative professionals to reduce repetitive or generic results, aiming for outputs that closely resemble real-world imagery.
Microsoft highlighted that MAI-Image-1 performs especially well in rendering photorealistic scenes, including complex lighting effects like bounce light and reflections. The company emphasized that the model’s combination of speed and quality allows users to visualize ideas quickly, iterate on them, and then transfer results into other tools for further refinement.
Although not yet available in Microsoft’s mainstream products, MAI-Image-1 made its debut on LMArena, an AI model leaderboard where users test and vote on different models. It currently holds the ninth position—a strong start for a newly released model. Microsoft has also introduced other proprietary models recently, including MAI-Voice-1 for natural speech generation and MAI-1-preview, a large-language model for text generation, both of which are also accessible via LMArena.
Microsoft stated that it is committed to ensuring safe and responsible AI outcomes, which is why it is gathering insights and feedback through public testing. The company plans to roll out MAI-Image-1 in Copilot and Bing Image Creator soon, but for now encourages users to try it on LMArena and share their impressions.
To test MAI-Image-1, visit the image generator page on LMArena. Use the drop-down menu at the top to switch from “Battle” mode to “Direct Chat.” Then select “mai-image-1” from the model menu, enter a descriptive prompt, and review the generated image.
You can also compare MAI-Image-1 with other image generators by selecting “Side by Side” from the same drop-down menu. Choose MAI-Image-1 as the first model and a competitor—such as DALL-E 3—as the second. Enter your prompt again to see side-by-side results and decide which output you prefer.
The collaboration between Microsoft and OpenAI was once very close, with Microsoft providing substantial funding and OpenAI supplying AI models that powered Microsoft products. Recently, however, the two companies have begun to pursue more independent paths. OpenAI is engaging with additional partners, while Microsoft is increasingly developing proprietary AI technology to support its expanding suite of services.
(Source: ZDNET)
