Iveda has added a new capability to its IvedaAI platform that allows security operators to create custom AI detection models using natural language prompts. The new capability is designed for rapid configuration of video surveillance in urban environments.

The “zero-shot” detection system enables users to type specific phrases, such as “suspicious behaviour” or “graffiti,” to instantly activate a detection model on live video feeds. This removes the traditional requirement for manual data labelling and weeks of model training.

The technology is designed for urban security and retail loss prevention. It utilises Vision Language Models (VLMs) to interpret the context of actions, such as an individual scanning staff-only entry points or attempting to conceal items. The system is currently being evaluated by a global fashion retailer with operations in over 90 countries.

Iveda offers two deployment methods: a cloud-connected mode for rapid activation and an on-premise server option called the Cosmos-Reason engine. The latter is intended for organisations requiring closed-network environments with no open ports to the internet.

“For years, building an AI detection model meant collecting data, labelling it, and waiting,” said David Ly, Chief Executive and Founder of Iveda. “Now, a retailer can type ‘shoplifting’ and, in seconds, have a fully functioning AI model running live.”