MAI-Image-1: After Gemini And ChatGPT, Microsoft Launches In-House AI Image Generator — How It Differs

Microsoft announced 'MAI-Image-1' its first-in house image generator via a blog post from the company on Tuesday.

It was launched amid the software giant's shift away from depending on OpenAI's Dall-E AI model for image generation in their large language model systems.

Its core strength, according to Microsoft, is its ability to generate photo-realistic images with a focus on realistic lighting, such as bouncing off and reflecting light in a believable manner.

It is also meant to avoid generating generic and repetitive images having been created with feedback from professional individuals in the creative industry.

"For example, we prioritized rigorous data selection and nuanced evaluation focused on tasks that closely mirror real-world creative use cases – taking into account feedback from professionals in the creative industries," the company said.

'MAI-Image 1' is ranked at number '9' in 'LMArena's' list of top 10 text-to-image AI generators, lagging behind Google's Gemini Nano-Banana and Imagen which hold the Number 1 and Number 3 spots, with Tencent's Hunyuan Image-3 sharing the top spot with Gemini.

Tencent's 'Seedream' also surpassed the image maker sitting at the Number 4 spot. ChatGPT's Gpt-image-1 generator was two spots below MAI at number 7.

They are ranked on their ability to generate images that match the descriptions given to the AI via text-based prompts.

Microsoft said that MAI works best when generating photo-realistic images with visual depth and clarity in comparison to "larger, slower" AI models. Which 'LMArena' might not have focused on, opting for a more holistic ranking.

"Its combination of speed and quality means users can get their ideas on screen faster, iterate through them quickly, and then transfer their work to other tools to continue refining," the company said.

The model is currently in its testing phase and results may vary as Microsoft refines its AI. It is expected to be available in Copilot and Bing Image Creator, "very soon" according to the company.

Users can also try the model for themselves in 'LMArena'.