Two years ago, Microsoft announced its Florence computer vision model as an AI system that was completely new to all existing models. Now Microsoft is making the model available for preview.

Unlike most vision models, Florence is both “unified” and “multimodal”. In concrete terms, this means that the model can understand both speech and images and can perform many different tasks.

Microsoft is currently working hard to commercialize its AI research. As part of this effort, Microsoft is adding Florence to an update to the Vision APIs in Azure Cognitive Services.

The Florence-based Microsoft Vision Services are now in preview for existing Azure customers. The AI ​​system allows users to generate automatic subtitles, remove backgrounds from images, create summaries of videos, and retrieve images, among other things.

What can you practically do with Florence?

A number of major Microsoft customers have already confirmed that they will actively use the new Microsoft Vision Services. For example, in an interview with TechCrunch, John Montgomery, CVP of Azure AI, confirmed that Reddit will use the service to generate alt text for images.

“Florence’s ability to generate up to 10,000 tags per image gives Reddit much more control over how many objects they can identify in an image and helps create much better descriptions,” Montgomery said.

Microsoft itself also uses Florence for its own products and services. For example, Microsoft will generate captions and alt text on LinkedIn using the Florence model. Microsoft Teams gets better video segmentation capabilities thanks to the AI ​​model. PowerPoint, Outlook, and Word also get features to automatically generate descriptions and alt text for images. Image tagging and image search capabilities are improved within Designer and OneDrive.