In an era where businesses are producing unprecedented amounts of video content, much of it remains untapped and unexamined. From extensive broadcast archives to countless hours of production footage, this dark data is a significant yet overlooked asset that companies rarely utilize effectively.
To address this challenge, Aza Kai (CEO) and Hiraku Yanagita (COO), both former Googlers with nearly a decade of collaboration at Google Japan, launched their own venture. They co-founded InfiniMind, a startup based in Tokyo, focused on developing infrastructure that transforms vast amounts of unviewed video and audio into organized, searchable business intelligence.
"While at Google, my co-founder and I recognized a pivotal moment approaching," Kai remarked. With advancements in technology and a clear market demand by 2024, they felt inspired to establish their own company.
Kai, who has a background in cloud services, machine learning, ad systems, and video recommendation models at Google, pointed out that existing solutions often force a compromise. Previous methods could identify objects in individual frames but struggled to track narratives, comprehend causality, or respond to complex inquiries about video content. This left many clients with extensive archives unable to answer even fundamental questions regarding their footage.
The landscape shifted significantly between 2021 and 2023 with advancements in vision-language models. Kai highlighted that video AI began evolving beyond mere object tagging during this period. The decline in GPU costs and annual performance improvements of approximately 15-20% over the last decade contributed to this evolution, but the real breakthrough lay in the capabilities of the models, which had previously been inadequate for the task.
Recently, InfiniMind secured $5.8 million in seed funding, led by UTEC, with participation from CX2, Headline Asia, Chiba Dojo, and an AI researcher at a16z Scout. The company plans to move its headquarters to the U.S. while maintaining an office in Japan, which provided an ideal environment for testing its technology with demanding clients.
In April 2025, InfiniMind launched its first product, TV Pulse, in Japan. This AI-driven platform analyzes television content in real-time, enabling media and retail businesses to monitor product exposure, brand presence, customer sentiment, and public relations impact. Following pilot programs with major broadcasters, it has already attracted paying customers, including wholesalers and media firms.
Now, InfiniMind is set to expand into international markets with its flagship product, DeepFrame, a comprehensive video intelligence platform designed to process 200 hours of footage to identify specific scenes, speakers, or events. The beta version is expected to be released in March, with a full launch slated for April 2026.
The video analysis sector is quite fragmented, with companies like TwelveLabs offering general-purpose video understanding APIs for a diverse user base. In contrast, InfiniMind zeroes in on enterprise applications, focusing on monitoring, safety, security, and in-depth video content analysis.
"Our solution is code-free; clients simply provide their data, and our system delivers actionable insights," Kai explained. "We also incorporate audio and speech understanding, not just visuals. Our platform can manage unlimited video lengths, and cost-effectiveness is a significant advantage. Many existing solutions prioritize accuracy or specific use cases but overlook cost-related challenges."
The seed funding will facilitate the ongoing development of the DeepFrame model, enhance engineering infrastructure, and support the recruitment of additional engineers to reach more customers in Japan and the U.S.
"This field is incredibly exciting and represents a pathway toward AGI," Kai stated. "Gaining insights into general video intelligence is vital for understanding reality. While industrial applications are crucial, our ultimate ambition is to push technological boundaries to enhance our comprehension of reality and assist humans in making informed decisions."
