“To understand the nature and movement of water, one must observe its waves.” This ancient wisdom, dating back more than 2,300 years, is at the heart of Guanlan, Hikvision’s newly launched family of large-scale AI models. The name isn’t just poetic – it reflects the company’s core philosophy: that true intelligence comes from perceiving subtle changes and uncovering deeper truths. In today’s fast-moving digital landscape, it’s a principle that resonates more than ever.
At its global launch event, Hikvision introduced Guanlan as a foundation for the AI-powered Internet of Things (AIoT) – a term the company uses to describe the fusion of artificial intelligence with sensor-rich IoT ecosystems. Since 2016, Hikvision has steadily expanded its AIoT capabilities beyond visible-light imaging, incorporating thermal, radar, millimetre wave, X-ray, and acoustic inputs into a comprehensive technology system. The arrival of Guanlan marks a significant leap forward in that journey, bringing unified, scalable AI capabilities into practical, high-impact deployments.
“The Guanlan models are not only built for perception, they are built for understanding,” said Darcy Yang from the Hikvision Research Institute during the presentation. “They aim to enhance the connection between the physical and digital world, promoting the development of intelligent societies, industries, and everyday lives.”
Deployment Flexibility for Real-World Use
One of the central strengths of the Guanlan models is their deployment versatility. Recognising that real-world applications demand a range of performance, latency, and privacy needs, Hikvision has designed a three-tier system to suit every scenario:
- Edge deployment brings AI directly to the front lines, with models running on devices like AI-enabled IP cameras and NVRs. This ensures low latency and preserves data privacy by avoiding unnecessary cloud transfer.
- Centre deployment moves processing to powerful server clusters, offering economies of scale and high-performance computation.
- Centre-edge fusion bridges the two approaches, allowing dynamic distribution of workloads based on the application context.
This flexibility enables Hikvision’s AIoT systems to operate optimally anytime, anywhere, whether it’s a remote construction site or busy main roads in the city centre. “The edge-centre fusion strategy ensures we don’t just deploy AI, we deploy the right AI in the right place,” explained Yang. “It’s how we meet low-latency demands without sacrificing depth or accuracy.”
Products Powered by Guanlan
Two flagship product lines powered by Guanlan were launched at the event, showcasing the practical reach of large-scale AI in security.
The AcuSeek NVRs transform video search from a time-consuming, manual task into a quick, intuitive experience. By combining large vision and multimodal models, these devices allow users to find specific events using natural language without needing tags, timestamps, or predefined filters. A search such as “a man in red clothes riding a bicycle” yields results in seconds.
“Security professionals have long been stuck in a race against time, scrubbing through hours of footage to find one key frame,” said Leo Chen, NVR Product Manager. “With Guanlan, we’ve moved from merely recording events to actively understanding them.”
Meanwhile, the new DeepinViewX cameras bring Guanlan’s large vision models to the edge, significantly improving perimeter protection. These intelligent cameras offer extended video content analysis (VCA) range, up to 120 metres for fixed models and 400 metres for PTZs, alongside drastically reduced false alarms, even in low-light or visually noisy environments.
“In low light and complex scenarios, our DeepinViewX cameras showed more than a 90% reduction in false alarms compared to conventional AI,” noted Bert Yao, Product Manager for the DeepinViewX line. “It’s not just about accuracy, it’s about trust.”
For customers with existing systems, Hikvision is also offering Guanlan-powered AI boxes and NVRs that can retrofit legacy installations. This makes it possible to add large-model intelligence without replacing every device on the network.
AIoT Technologies Driven by Guanlan
At its core, Guanlan is designed to push the boundaries of AIoT by embedding intelligence directly into the sensors, systems, and workflows of diverse industries and scenarios. These models enable advanced analysis of images, video, voice, and environmental data to identify patterns, classify behaviours, and predict outcomes. From cutting-edge video analytics to ever-expanding dimensions of perception, Guanlan serves as the cornerstone for smarter, faster, and clearer insights at lower costs.
In real-world applications, the roadmap includes open-vocabulary object detection, automatic number plate recognition, video structuring, and real-time activity monitoring, functions that apply just as easily to traffic management or industrial safety as to traditional security.
Unlike standard deep learning models, Guanlan leverages a transformer-based architecture and self-supervised pretraining, while incorporating rich industry insights, allowing it to generalise across varied scenarios while remaining sensitive to local context.
Looking Forward
As the industry enters the age of large-scale AI, the Guanlan models stand out for their real-world grounding. This isn’t AI for its own sake. It’s AI for safer campuses, more responsive emergency systems, smarter cities, and more resilient operations. It’s about intelligence that adapts, anticipates, and assists on-site, on time, and in context.
“In the past, people searched through video like finding a needle in a haystack,” Yang reflected. “Now, machines can understand what you want and find it for you in seconds. This is not the future. This is now.”
Find out more here.