Gemini can now natively embed video, so I built sub-second video search

Name: Gemini can now natively embed video, so I built sub-second video search
Availability: InStock
Author: sohamrj

by sohamrj·Mar 24, 2026·438 points·108 comments

Visit Project View on HN

AI Analysis

●●●BangerWizardryZero to OneBig Brain

Direct video-to-vector embedding skips transcription entirely—Twelve Labs but self-hosted.

Strengths

•Gemini's native video embedding avoids transcription errors and captures visual context
•Still-frame detection skips idle chunks, cutting costs for security camera footage
•Working CLI with auto-trim output solves a genuinely painful search problem

Weaknesses

•Requires Gemini API key; no offline embedding option for sensitive footage
•Dashcam niche limits broader adoption beyond security and fleet use cases

Post Description

Gemini Embedding 2 can project raw video directly into a 768-dimensional vector space alongside text. No transcription, no frame captioning, no intermediate text. A query like "green car cutting me off" is directly comparable to a 30-second video clip at the vector level.

I used this to build a CLI that indexes hours of footage into ChromaDB, then searches it with natural language and auto-trims the matching clip. Demo video on the GitHub README. Indexing costs ~$2.50/hr of footage. Still-frame detection skips idle chunks, so security camera / sentry mode footage is much cheaper.