Azure Vision in Foundry Tools

Azure Vision is a powerful tool within Foundry Tools (formerly Azure AI Services). Azure Vision provides a set of prebuilt APIs that enable applications and agents to visually interpret the world. Vision provides capabilities such as image analysis, object detection, spatial understanding, and optical character recognition (OCR). These tools allow developers to build intelligent solutions that can "see" and understand visual content in real time.

Vision is integrated into the unified Foundry platform to deliver advanced agentic AI experiences.

Azure Vision (formerly Azure AI Vision) is designed to accelerate the development of intelligent agents and applications without requiring deep machine learning expertise.

Yes, Azure Vision (formerly Azure AI Vision) is now part of the Foundry Tools suite. This rebranding is part of a broader platform unification under Foundry, designed to reflect how developers are increasingly using these services as modular tools to build intelligent, agentic applications.

Azure Vision in Foundry Tools continues to offer the same powerful capabilities—like image analysis, object detection, OCR, and spatial understanding—but is now positioned within a cohesive toolkit that supports agent workflows and multimodal AI scenarios. The rebrand helps clarify how Vision fits into the Foundry ecosystem, making it easier to discover, orchestrate, and integrate with other agents and tools.

This shift is about empowering developers with a unified experience for building AI agents that see, understand, and act.

No. Microsoft automatically deletes your images and videos after processing and does not train on your data to enhance the underlying models. Video data does not leave your premises, and video data is not stored on the edge where the container runs. Learn more about privacy and terms of usage.

No, spatial analysis detects and locates human presence in video footage and outputs a bounding box around each person detected. The AI models do not detect faces nor determine individuals’ identities nor demographics.

The spatial analysis AI models detect and track movements in the video feed based on algorithms that identify the presence of one or more humans by a body bounding box. For each person and bounding box detected in a zone in the camera field of view, the AI models output event data including bounding box coordinates of a person’s body, event type (for example, zone entry or exit, or directional line crossing), pseudonymous identifiers to track the bounding box, and a detection confidence score. This event data is sent to your own instance of Azure IoT Hub.

Yes. Because model customization is designed to be fine-tuned for your scenario, you need to provide labeled data to train your model.

The model customization feature of the service is optimized to quickly recognize major differences between images, so you can start prototyping your model with a small amount of data. You may start with as little as one image per label. If you have more labeled images, you may add more. Depending on the complexity of the problem and degree of accuracy required, you can continue adding additional images per label to improve your model.

The model customization feature for Azure Vision is the next generation of Custom Vision, with improved accuracy and few-shot learning capabilities.It is recommended that you migrate your training data to retrain your model with model customization in Azure Vision.