Iveda® (NASDAQ: IVDA) announced that its IvedaAI platform now supports real‑time zero‑shot AI detection driven by natural‑language prompts. Retailers can type a single word such as “shoplifting” or “suspicious behavior” and receive an instant, custom detection model that runs on live video feeds without any prior data labeling or model training. The capability is positioned as a faster, more flexible way to address loss‑prevention and other security concerns across large camera networks.
Real‑Time Zero‑Shot Detection Launched on IvedaAI
The new feature lets users create a detection model in seconds by entering a prompt. IvedaAI then builds a custom model on the fly and applies it to live or recorded video streams. No separate login, infrastructure change, or deployment lag is required.
David Ly, CEO and Founder of Iveda, said the advance “is the greatest leap forward yet in real‑time AI video analytics,” noting that retailers can now type “shoplifting” and have a fully functioning AI model running across their camera network within seconds. The company highlighted early testing where a prompt of “suspicious behavior” successfully flagged individuals peeking into windows and repeatedly scanning staff‑only entry points.
Targeted Retail Use Cases and Early Evaluation
IvedaAI’s zero‑shot approach is being evaluated with one of the world’s largest fast‑fashion retailers, which operates thousands of stores in more than 90 countries. The platform now supports live prompts for:
- Shoplifting across diverse concealment methods
- Suspicious or pre‑incident behavior
- Loitering in sensitive areas
- Graffiti attempts and vandalism indicators
- Smoking in restricted zones
- Fighting or aggressive conduct
All detections appear directly in the retailer’s existing IvedaAI dashboard. The company emphasized that traditional AI models struggle with the variability of shoplifting tactics, whereas the vision‑language model (VLM) reasoning can interpret intent and context rather than relying solely on static object detection.
Deployment Options: Cloud‑Connected or Fully On‑Premise
IvedaAI offers two pathways for the zero‑shot capability:
- Cloud‑Connected Mode – Users connect to large‑language‑model processing for live frame analysis, avoiding additional on‑site hardware.
- Local Secure Deployment (Cosmos‑Reason Engine) – For closed‑network environments, Iveda provides a dedicated on‑premise server powered by its proprietary Cosmos‑Reason engine, ensuring zero open ports and fully local processing. The next‑generation Cosmos‑Reason‑2 engine is slated to ship within the coming quarter, promising higher inference performance and detection accuracy.
Availability, Pricing and Platform Foundations
The zero‑shot detection is available now as an add‑on annual license within the IvedaAI platform. Existing customers can activate it through their current secure camera monitoring dashboard. Pricing details are hosted at iveda.com/ivedaai.
Iveda notes that the capability builds on more than a decade of AI object‑detection work, including weapons recognition, license‑plate identification, and facial recognition. The company also mentioned ongoing integration of the latest NVIDIA frameworks to keep the platform current.
Key Takeaways
- IvedaAI now supports real‑time zero‑shot detection that creates a custom model from a single natural‑language prompt in seconds, with no prior data labeling or training.
- The feature is being evaluated with a major fast‑fashion retailer operating thousands of stores in over 90 countries, and it currently supports prompts for shoplifting, suspicious behavior, loitering, graffiti, smoking, and fighting.
- Deployment is offered via a cloud‑connected mode or a fully on‑premise Cosmos‑Reason engine, with the next‑generation Cosmos‑Reason‑2 hardware expected to ship in the next quarter.
TechInsyte's Take
Iveda’s prompt‑driven detection could reduce the time and expertise needed for retailers to add new security use cases, making AI surveillance more adaptable to evolving loss‑prevention threats. Buyers should monitor the performance of the upcoming Cosmos‑Reason‑2 hardware and assess how the cloud versus on‑premise options align with their data‑privacy policies. The real‑world accuracy and false‑positive rates of the zero‑shot models remain to be validated at scale.
Source: Businesswire
