Show HN: Gemini can now natively embed video, so I built sub-second video search (github.com)

Article summary

The author has built a tool called SentrySearch that uses Gemini's video embedding API to enable sub-second video search. The tool splits videos into chunks, embeds each chunk as a vector, and stores them in a database, allowing for fast search and retrieval of specific clips. The tool also supports local models for private and offline search. The author provides examples of use cases, including dashcam and home security footage search.

Main themes

• Video Search
• Gemini Embedding API
• Local Models
• Surveillance
• Privacy
• AI Applications

What commenters say

• The tool has potential applications in various fields, including surveillance and home security, but raises concerns about privacy and potential misuse.
• The use of AI-powered video search could lead to a dystopian future where every movement is tracked and analyzed.
• The technology is not yet practical for large-scale use due to costs and limitations, but could become more viable with advancements in embedding models.
• Some commenters see the tool as a useful solution for searching and analyzing video footage, while others are concerned about the potential for abuse and erosion of privacy.
• The development of open-weight models for video embedding could eliminate API costs and privacy concerns, making the technology more accessible and widely adopted.
• The tool's ability to search and analyze video footage could have significant implications for fields such as marketing, branding, and social media monitoring.