InfiniMind raises 5.8 million dollars to turn “dark” enterprise video into queryable data

Companies are sitting on a growing pile of “dark data”: years of broadcast archives, store cameras, and production footage that is stored but rarely searched, measured, or monetized. InfiniMind is betting that AI can turn that backlog into structured business data marketers and operators can actually query.

Enterprise video AI is shifting from tagging to narrative understanding

InfiniMind, founded by former Google Japan colleagues Aza Kai (CEO) and Hiraku Yanagita (COO), is building infrastructure that converts petabytes of video and audio into searchable, analysis-ready outputs. The pitch is less “object detection in frames” and more end-to-end understanding: tracking who said what, when a product appears, and how context changes across long-form footage.

This capability jump is tied to advances in vision-language models from 2021 to 2023, which moved the industry beyond basic labeling toward multi-modal reasoning (combining visuals with speech and sound). The team argues that earlier systems forced a trade-off: they could tag items, but struggled with causality, storylines, and complex queries across large archives.

The market backdrop is simple: businesses are producing more video, and forecasts suggest video will dominate consumer internet traffic by 2025, making the analysis gap more expensive over time (Personify). Meanwhile, unused operational content is increasingly recognized as “dark data” with unrealized business value (iconik).

TV Pulse is live in Japan, DeepFrame targets long-form search in 2026

The startup raised 5.8 million dollars in seed funding led by UTEC, with participation from CX2, Headline Asia, Chiba Dojo, and an AI researcher via a16z Scout. InfiniMind is relocating its headquarters to the US while maintaining a Japan office.

Product-wise, TV Pulse launched in Japan in April 2025, aimed at real-time TV analysis for media and retail teams tracking product exposure, brand presence, sentiment, and PR impact. Next is DeepFrame, positioned as a long-form video intelligence platform that can process about 200 hours of footage to find specific scenes, speakers, or events, with beta planned for March and a full release in April 2026. For B2B marketers, the practical upside is faster measurement of brand visibility and sponsorship outcomes across large video libraries, without manual logging.

InfiniMind’s bet: enterprise-grade, no-code video intelligence will be won on cost efficiency and multi-modal depth, not just benchmark accuracy.