VK Video boosts views by recognizing favorite characters in clips

VK has rolled out advanced content discovery technology that automatically identifies well-known personalities on its video platform. Since implementing this, the average watch time for clips featuring the same characters in the “Watch Also” section has increased by 9%.

This recognition system relies on two machine learning models working together. The first scans video frames at one-per-second intervals to detect any people on screen. The second cross-references those detected faces against a database of known personalities, pinpointing familiar figures. This means VK’s recommendation engine now considers not just user behavior, topics, and video content, but also the specific characters appearing in videos.

Beyond character recognition, VK’s Discovery platform includes two additional AI components: a multimodal language model that interprets video narratives, and a cross-format content model that understands the meaning of content regardless of its form. VK’s AI engineers unveiled these innovations at the end of 2025.

Recommended reading

Claude Code’s 60-Second Auto-Continue Backfired

VK Video’s approach contrasts with Western platforms like YouTube and TikTok, which primarily rely on metadata, user interaction patterns, and thumbnail analysis for recommendations. By embedding direct recognition of known figures within videos, VK aims to deepen personalized viewing experiences and keep audiences engaged longer-especially in Russia’s evolving digital landscape, where local content drives viewership.

Looking ahead, the key questions are how scalable this technology is beyond Russian content and whether other streaming platforms will adopt similar character-aware recommendation systems. If VK’s AI proves effective, it could set a new standard for video recommendations that connect audiences more closely with their favorite on-screen personalities.