Hikvision upgrades Guanlan AI

Hikvision said on Tuesday that it has upgraded applications built on its Guanlan large-scale AI models, adding more accurate search, lower-cost video handling and natural-language interaction for surveillance systems. The Hangzhou-based company said the software can identify what appears in live or recorded footage and return results through language-based queries. Hikvision framed the update as the next step after launching Guanlan more than a year ago. The company’s materials describe Guanlan as a stack of vision, language and multimodal models for its AIoT products. ### What is Hikvision saying has changed in the product? Hikvision said the new applications allow video systems to “understand” scenes, recognize what has already been recorded and respond to user requests through natural-language interaction. In its September 8 release, the company said those capabilities are intended to reduce manual review and improve fit for specific deployment scenarios. (prnewswire.com) The company has described Guanlan since its 2025 launch as a three-tier architecture built from foundation models, industry models and task models. Hikvision said at that launch that the system combined large vision, language and multimodal models to support AI-powered Internet of Things products and applications. (prnewswire.com) ### How does natural-language search change a video system in practice? Recorded video has usually been searched through timestamps, motion flags or predefined analytics rules. Hikvision said Guanlan adds language-based interaction, allowing users to ask for events or objects in ordinary text rather than building a query through menus and filters. The company said that lets systems work with what a user wants and what the system has already captured. (prnewswire.com) Hikvision’s own product pages say Guanlan is meant to push AIoT devices toward “smarter, faster, and clearer insights” by embedding advanced AI engines across products and applications. That description does not give an independent benchmark, but it shows the company is positioning Guanlan as a common intelligence layer rather than a single camera feature. ### Where does the cost claim come from? (prnewswire.com) Hikvision tied part of the lower-cost pitch to storage and compression. In a May 21 release, the company said its Guanlan Encoding video compression technology delivers average storage savings of 30% to 50% without compromising critical footage quality. That release predates Tuesday’s announcement, but it is the clearest numerical cost figure attached to the Guanlan product line. (hikvision.com) Marketscreener’s summary of Hikvision’s earlier Guanlan launch also said DeepinViewX cameras using the model cut false alarms by 90%, doubled detection range and halved repeated alarms on the same target, citing company claims. Those figures relate to a separate launch, but they show the types of performance gains Hikvision has used to market Guanlan-based products. (prnewswire.com) ### Which workflows could this affect for security operators? Industrial parks, campuses and other large sites typically generate long stretches of retained footage and large volumes of alerts. Hikvision said Guanlan applications are meant to lower manual work and make operations safer and less costly by matching the software more closely to the scenario where it is deployed. (marketscreener.com) A system that can search footage semantically may change how operators review incidents, export clips and decide what video to keep at full quality. Hikvision did not set out new retention rules or compliance policies in Tuesday’s release, and any changes to those practices would depend on the operator, the site and local legal requirements. (prnewswire.com) ### What comes next from Hikvision’s Guanlan rollout? Hikvision’s September 8 release is the latest public update in a sequence that included the original Guanlan launch in 2025 and the Guanlan Encoding release in May 2026. The company said the models are being integrated across products and applications, and its public Guanlan pages continue to position the technology as a core layer in future AIoT deployments. (prnewswire.com)